diff --git a/.github/dependabot.yml b/.github/dependabot.yml
new file mode 100644
index 0000000000000000000000000000000000000000..276690bd4f96f50eb6a5121d237c8c9bcb80224d
--- /dev/null
+++ b/.github/dependabot.yml
@@ -0,0 +1,12 @@
+# To get started with Dependabot version updates, you'll need to specify which
+# package ecosystems to update and where the package manifests are located.
+# Please see the documentation for all configuration options:
+# https://docs.github.com/github/administering-a-repository/configuration-options-for-dependency-updates
+
+version: 2
+updates:
+  - package-ecosystem: "pip" # See documentation for possible values
+    directory: "/docs/sphinx" # Location of package manifests
+    open-pull-requests-limit: 10
+    schedule:
+      interval: "daily"
diff --git a/.gitignore b/.gitignore
index 71059ec4d948ed241b76ea7ebf9c79136cfcbf74..7af066c82dcef3cf4cc846168567e90e0a0c9db3 100644
--- a/.gitignore
+++ b/.gitignore
@@ -47,3 +47,12 @@ build*
 # GDB temporary files
 .gdb_history
 install.dir*
+
+# documentation artifacts
+_build/
+_images/
+_static/
+_templates/
+_toc.yml
+docBin/
+_doxygen/
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..3e1e8c0169fc7875a0795db3a9c9d0149b40eb87
--- /dev/null
+++ b/.pre-commit-config.yaml
@@ -0,0 +1,14 @@
+repos:
+-   repo: local
+    hooks:
+    -   id: clang-format
+        name: clang-format
+        entry: clang-format-10 -i --style=file
+        language: system
+        types_or: [c++, inc]
+    -   id: copyright-year-checker
+        name: copyright-year-checker
+        entry: script/check_copyright_year.sh
+        verbose: false
+        language: script
+        types: [c++]
diff --git a/.readthedocs.yaml b/.readthedocs.yaml
new file mode 100644
index 0000000000000000000000000000000000000000..5f50df2525d2948bbabe36184f23310d47339d8e
--- /dev/null
+++ b/.readthedocs.yaml
@@ -0,0 +1,18 @@
+# Read the Docs configuration file
+# See https://docs.readthedocs.io/en/stable/config-file/v2.html for details
+
+version: 2
+
+build:
+   os: ubuntu-22.04
+   tools:
+      python: "3.8"
+
+sphinx:
+   configuration: docs/conf.py
+
+formats: [htmlzip, pdf, epub]
+
+python:
+   install:
+   - requirements: docs/sphinx/requirements.txt
diff --git a/CHANGELOG.md b/CHANGELOG.md
new file mode 100644
index 0000000000000000000000000000000000000000..3898b5ce2d9c1a4006209a1aa9fa6e48302ea30a
--- /dev/null
+++ b/CHANGELOG.md
@@ -0,0 +1,27 @@
+# Change Log for Composable Kernel
+
+Full documentation for Composable Kernel is not yet available.
+
+## CK 0.2.0 for ROCm 5.5.0
+
+### Fixed
+- Fixed a bug in 6-dimensional kernels (#555).
+- Fixed grouped ConvBwdWeight test case failure (#524).
+
+### Optimizations
+- Improve proformance of normalization kernel
+
+### Added
+- Added support on NAVI3x.
+- Added user tutorial (#563).
+- Added more instances for irregular GEMM sizes (#560).
+- Added inter-wave consumer-producer programming model for GEMM kernels (#310).
+- Added multi-D GEMM client APIs (#534).
+- Added multi-embeddings support (#542).
+- Added Navi3x blockwise GEMM and real GEMM support (#541).
+- Added Navi grouped ConvBwdWeight support (#505).
+- Added pool3d forward (#697).
+- Added maxpool backward (#750).
+
+### Changed
+- Changed ...
diff --git a/CMakeLists.txt b/CMakeLists.txt
index f861e302039de8e2c1ae17bf7cb39361aff8535f..c9fb6b4552c81ef29e77dcde807f552b42916572 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -22,6 +22,7 @@ include(TargetFlags)
 list(APPEND CMAKE_PREFIX_PATH ${CMAKE_INSTALL_PREFIX} ${CMAKE_INSTALL_PREFIX}/llvm ${CMAKE_INSTALL_PREFIX}/hip /opt/rocm /opt/rocm/llvm /opt/rocm/hip)
 
 option(USE_BITINT_EXTENSION_INT4, "Whether to enable clang's BitInt extension to provide int4 data type." OFF)
+option(USE_OPT_NAVI3X, "Whether to enable LDS cumode and Wavefront32 mode for NAVI3X silicons." OFF)
 
 if(USE_BITINT_EXTENSION_INT4)
     add_compile_definitions(CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4)
@@ -29,6 +30,12 @@ if(USE_BITINT_EXTENSION_INT4)
     message("CK compiled with USE_BITINT_EXTENSION_INT4 set to ${USE_BITINT_EXTENSION_INT4}")
 endif()
 
+if(USE_OPT_NAVI3X)
+    add_compile_options(-mcumode)
+    add_compile_options(-mno-wavefrontsize64)
+    message("CK compiled with USE_OPT_NAVI3X set to ${USE_OPT_NAVI3X}")
+endif()
+
 ## Threads
 set(THREADS_PREFER_PTHREAD_FLAG ON)
 find_package(Threads REQUIRED)
diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
index 8ccfe99c3cc73b643f8b92cb654005e54c0774bd..07d83688176bd771142a8698fda492c538c691b1 100644
--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -4,7 +4,7 @@ This is the list of developers and contributors to Composable Kernel library
 
 
 ## Developers
-[Chao Liu](https://github.com/asroy), [Jing Zhang](https://github.com/zjing14), 2018-2022
+[Chao Liu](https://github.com/asroy), [Jing Zhang](https://github.com/zjing14), 2018-2023
 
 [Letao Qin](https://github.com/ltqin), [Qianfeng Zhang](https://github.com/qianfengz), [Liang Huang](https://github.com/carlushuang), [Shaojie Wang](https://github.com/shaojiewang), 2019-2022
 
diff --git a/Dockerfile b/Dockerfile
index d024f966c57fac8ced14abbb00db7cc8f20c7d63..3ff29c9a98a56cb1c26bb8e895f336d4fca6475b 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -1,28 +1,44 @@
 FROM ubuntu:20.04
-
-ARG ROCMVERSION=5.3
-ARG compiler_version="release"
+ARG DEBIAN_FRONTEND=noninteractive
+ARG ROCMVERSION=5.6
+ARG compiler_version=""
 ARG compiler_commit=""
 
 RUN set -xe
 
 ARG DEB_ROCM_REPO=http://repo.radeon.com/rocm/apt/.apt_$ROCMVERSION/
+RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
 # Add rocm repository
+RUN chmod 1777 /tmp
 RUN apt-get update
-RUN apt-get install -y wget gnupg
-RUN wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add -
-RUN sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"
-RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
+RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
+
+ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
+RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
+
+RUN wget https://repo.radeon.com/amdgpu-install/5.6/ubuntu/focal/amdgpu-install_5.6.50600-1_all.deb  --no-check-certificate
+RUN apt-get update && \
+DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
+    ./amdgpu-install_5.6.50600-1_all.deb
+
+RUN if [ "$ROCMVERSION" != "5.7" ]; then \
+        wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
+        sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
+        sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
+    elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
+         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
+         apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
+         amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
+    fi
+
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
+RUN amdgpu-install -y --usecase=rocm --no-dkms
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
-    apt-utils \
     build-essential \
     ccache \
-    cmake-data \
     cmake \
-    curl \
     git \
     hip-rocclr \
     jq \
@@ -32,16 +48,13 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     libpthread-stubs0-dev \
     llvm-amdgpu \
     pkg-config \
-    python \
     python3 \
-    python-dev \
     python3-dev \
     python3-pip \
+    sshpass \
     software-properties-common \
-    rocm-dev \
-    rocm-device-libs \
-    rocm-cmake \
     vim \
+    nano \
     zlib1g-dev \
     openssh-server \
     clang-format-10 \
@@ -49,6 +62,17 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+#Install latest version of cmake
+RUN apt purge --auto-remove -y cmake
+RUN apt update
+RUN apt install -y software-properties-common lsb-release
+RUN apt clean all
+RUN wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | gpg --dearmor - | tee /etc/apt/trusted.gpg.d/kitware.gpg >/dev/null
+RUN apt-add-repository "deb https://apt.kitware.com/ubuntu/ $(lsb_release -cs) main"
+RUN apt install -y kitware-archive-keyring
+RUN rm /etc/apt/trusted.gpg.d/kitware.gpg
+RUN apt install -y cmake
+
 # Setup ubsan environment to printstacktrace
 RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer
 ENV UBSAN_OPTIONS=print_stacktrace=1
@@ -60,7 +84,7 @@ RUN dpkg -i dumb-init_*.deb && rm dumb-init_*.deb
 ARG PREFIX=/opt/rocm
 # Install packages for processing the performance results
 RUN pip3 install --upgrade pip
-RUN pip3 install sqlalchemy
+RUN pip3 install sqlalchemy==1.4.46
 RUN pip3 install pymysql
 RUN pip3 install pandas
 RUN pip3 install setuptools-rust
@@ -84,12 +108,7 @@ ENV compiler_commit=$compiler_commit
 RUN sh -c "echo compiler version = '$compiler_version'"
 RUN sh -c "echo compiler commit = '$compiler_commit'"
 
-RUN --mount=type=ssh if [ "$compiler_version" = "amd-stg-open" ]; then \
-        sed -i '/$HIP_CLANG_TARGET = chomp($HIP_CLANG_TARGET);/c\    chomp($HIP_CLANG_TARGET);' /opt/rocm/hip/bin/hipcc.pl && \
-        sed -i '/$HIP_CLANG_TARGET = chomp($HIP_CLANG_TARGET);/c\    chomp($HIP_CLANG_TARGET);' /opt/rocm/bin/hipcc.pl; \
-    fi
-
-RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_commit" = "" ]; then \
+RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && mkdir build && cd build && \
         cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \
@@ -97,7 +116,7 @@ RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_com
     else echo "using the release compiler"; \
     fi
 
-RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_commit" != "" ]; then \
+RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
         cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \
diff --git a/Jenkinsfile b/Jenkinsfile
index 7b2e57c1403df3cd4574bf127b96163f1b736d17..df9d13167f56f69e77814979c3400ea3e03a1c5b 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -11,15 +11,56 @@ def show_node_info() {
     """
 }
 
+def nthreads() {
+    def nproc = sh(returnStdout: true, script: 'nproc')
+    echo "Number of cores: ${nproc}"
+    def n = nproc.toInteger()
+    if (n > 32){
+        n /= 2
+    }
+    if (n > 64){
+        n = 64
+    }
+    echo "Number of threads used for building: ${n}"
+    return n
+}
+
 def runShell(String command){
     def responseCode = sh returnStatus: true, script: "${command} > tmp.txt"
     def output = readFile(file: "tmp.txt")
-    echo "tmp.txt contents: $output"
     return (output != "")
 }
 
 def getDockerImageName(){
-    def img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}"
+    def img
+    if (params.ROCMVERSION != "5.7"){
+       if (params.COMPILER_VERSION == "") {
+           img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
+       }
+       else{
+          if (params.COMPILER_COMMIT == ""){
+             img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}"
+          }
+          else{
+             def commit = "${params.COMPILER_COMMIT}"[0..6]
+             img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}_${commit}"
+          }
+       }
+    }
+    else{
+       if (params.COMPILER_VERSION == "") {
+           img = "${env.CK_DOCKERHUB_PRIVATE}:ck_ub20.04_rocm${params.ROCMVERSION}"
+       }
+       else{
+          if (params.COMPILER_COMMIT == ""){
+             img = "${env.CK_DOCKERHUB_PRIVATE}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}"
+          }
+          else{
+             def commit = "${params.COMPILER_COMMIT}"[0..6]
+             img = "${env.CK_DOCKERHUB_PRIVATE}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}_${commit}"
+          }
+       }
+    }
     return img
 }
 
@@ -43,11 +84,11 @@ def build_compiler(){
         compiler = '/opt/rocm/bin/hipcc'
     }
     else{
-        if (params.COMPILER_VERSION == "release"){
-            compiler = "/opt/rocm/llvm/bin/clang++"
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
+            compiler = "/llvm-project/build/bin/clang++"
         }
         else{
-            compiler = "/llvm-project/build/bin/clang++"
+            compiler = "/opt/rocm/llvm/bin/clang++"
         }        
     }
     return compiler
@@ -165,7 +206,7 @@ def cmake_build(Map conf=[:]){
     if(conf.get("build_install","") == "true")
     {
         config_targets = 'install ' + config_targets
-        setup_args = ' -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install' + setup_args
+        setup_args = ' -DBUILD_DEV=On -DCMAKE_INSTALL_PREFIX=../install' + setup_args
     } else{
         setup_args = ' -DBUILD_DEV=On' + setup_args
     }
@@ -192,7 +233,8 @@ def cmake_build(Map conf=[:]){
         """
     def setup_cmd = conf.get("setup_cmd", "${cmake_envs} cmake ${setup_args}   .. ")
     // reduce parallelism when compiling, clang uses too much memory
-    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j\$(( \$(nproc) / 2 )) ${config_targets}")
+    def nt = nthreads()
+    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j${nt} ${config_targets}")
     def execute_cmd = conf.get("execute_cmd", "")
 
     def cmd = conf.get("cmd", """
@@ -226,7 +268,7 @@ def buildHipClangJob(Map conf=[:]){
             dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
         }
         def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-        if (params.COMPILER_VERSION != "release"){
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
         }
 
@@ -281,7 +323,7 @@ def runCKProfiler(Map conf=[:]){
             dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
         }
         def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-        if (params.COMPILER_VERSION != "release"){
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
         }
 
@@ -414,12 +456,13 @@ def Build_CK(Map conf=[:]){
             dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
         }
         def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-        if (params.COMPILER_VERSION != "release"){
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
         }
 
         def variant = env.STAGE_NAME
         def retimage
+        def navi_node = 0
 
         gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCmSoftwarePlatform', repo: 'composable_kernel') {
             try {
@@ -433,6 +476,9 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
+                            navi_node = 1
+                        }
                     }
                 }
             }
@@ -451,6 +497,9 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
+                            navi_node = 1
+                        }
                     }
                 }
             }
@@ -460,10 +509,22 @@ def Build_CK(Map conf=[:]){
                     cmake_build(conf)
                     dir("build"){
                         //run tests and examples
-                        sh 'make -j check'
-                        //we only need the ckProfiler to run the performance tests, so we pack and stash it
-                        sh 'tar -zcvf ckProfiler.tar.gz bin/ckProfiler'
-                        stash "ckProfiler.tar.gz"
+                        def nt = nthreads()	
+                        sh 'make -j${nt} check'
+                        if (navi_node == 0 ){
+                            //we only need the ckProfiler to run the performance tests, so we pack and stash it
+                            //do not stash profiler on Navi nodes
+                           sh 'tar -zcvf ckProfiler.tar.gz bin/ckProfiler'
+                           stash "ckProfiler.tar.gz"
+                        }
+                        if (params.RUN_FULL_QA){
+                           // build deb packages
+                           sh 'make -j package'
+                           archiveArtifacts artifacts: 'composablekernel-ckprofiler_*.deb'
+                           archiveArtifacts artifacts: 'composablekernel-tests_*.deb'
+                           sh 'mv composablekernel-ckprofiler_*.deb ckprofiler_0.2.0_amd64.deb'
+                           stash "ckprofiler_0.2.0_amd64.deb"
+                        }
                     }
                 }
             }
@@ -530,6 +591,8 @@ def process_results(Map conf=[:]){
                         unstash "perf_splitK_gemm.log"
                         unstash "perf_onnx_gemm.log"
                         sh "./process_qa_data.sh"
+                        unstash "ckprofiler_0.2.0_amd64.deb"
+                        sh "sshpass -p ${env.ck_deb_pw} scp -o StrictHostKeyChecking=no ckprofiler_0.2.0_amd64.deb ${env.ck_deb_user}@${env.ck_deb_ip}:/var/www/html/composable_kernel/"
                     }
                     else{
                         // unstash perf files to master
@@ -550,8 +613,9 @@ def process_results(Map conf=[:]){
 }
 
 //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
-CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;COMPILER_VERSION=release
-                                              0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open''' : ""
+CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true
+                                              0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT=
+                                              0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 
 pipeline {
     agent none
@@ -568,16 +632,16 @@ pipeline {
             description: "Force building docker image (default: false), set to true if docker image needs to be updated.")
         string(
             name: 'ROCMVERSION', 
-            defaultValue: '5.3', 
-            description: 'Specify which ROCM version to use: 5.2.3, or 5.3 (default), etc.')
+            defaultValue: '5.6', 
+            description: 'Specify which ROCM version to use: 5.6 (default).')
         string(
             name: 'COMPILER_VERSION', 
-            defaultValue: 'release', 
-            description: 'Specify which version of compiler to use: ck-9110, release (default), or amd-stg-open.')
+            defaultValue: '', 
+            description: 'Specify which version of compiler to use: release, amd-stg-open, or leave blank (default).')
         string(
             name: 'COMPILER_COMMIT', 
             defaultValue: '', 
-            description: 'Specify which commit of compiler branch to use: leave empty to use the latest commit (default), or use 8a82e4eb7ba28521ba9a9424a0315a8a16590424 commit of amd-stg-open branch.')
+            description: 'Specify which commit of compiler branch to use: leave blank to use the latest commit, or use 5541927df00eabd6a110180170eca7785d436ee3 (default) commit of amd-stg-open branch.')
         string(
             name: 'BUILD_COMPILER', 
             defaultValue: 'hipcc', 
@@ -639,12 +703,47 @@ pipeline {
         {
             parallel
             {
-                stage("Build CK and run Tests")
+                stage("Build CK and run Tests on MI100/MI200/MI300")
                 {
+                    when {
+                        beforeAgent true
+                        expression { params.RUN_FULL_QA.toBoolean() }
+                    }
                     agent{ label rocmnode("gfx908 || gfx90a") }
                     environment{
-                        setup_args = "${params.COMPILER_VERSION == "ck-9110" ? """ -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a" -DCMAKE_CXX_FLAGS="-O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" """ : """ -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a" -DCMAKE_CXX_FLAGS="-O3 " """ }"
-                        execute_args = "${params.COMPILER_VERSION == "ck-9110" ? """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a" -DCMAKE_CXX_FLAGS="-O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ : """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908,gfx90a" -DCMAKE_CXX_FLAGS="-O3" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ }"
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                    }
+                    steps{
+                        Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                    }
+                }
+                stage("Build CK and run Tests on MI100/MI200")
+                {
+                    when {
+                        beforeAgent true
+                        expression { !params.RUN_FULL_QA.toBoolean() }
+                    }
+                    agent{ label rocmnode("gfx908 || gfx90a") }
+                    environment{
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                    }
+                    steps{
+                        Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                    }
+                }
+                stage("Build CK and run Tests on Navi21")
+                {
+                    when {
+                        beforeAgent true
+                        expression { !params.RUN_FULL_QA.toBoolean() }
+                    }
+                    agent{ label rocmnode("navi21") }
+                    environment{
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1030" """ 
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1030" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
+
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
@@ -657,7 +756,7 @@ pipeline {
         {
             parallel
             {
-                stage("Run ckProfiler: gfx908 or gfx90a")
+                stage("Run ckProfiler: gfx90*")
                 {
                     when {
                         beforeAgent true
@@ -666,7 +765,7 @@ pipeline {
                     options { retry(2) }
                     agent{ label rocmnode("gfx908 || gfx90a")}
                     environment{
-                        setup_args = "${params.COMPILER_VERSION == "ck-9110" ? """ -DGPU_TARGETS="gfx908;gfx90a" -DCMAKE_CXX_FLAGS=" -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -DBUILD_DEV=On """ : """ -DGPU_TARGETS="gfx908;gfx90a" -DCMAKE_CXX_FLAGS=" -O3 " -DBUILD_DEV=On """}"
+                        setup_args = """ -DGPU_TARGETS="gfx908;gfx90a" -DBUILD_DEV=On """
                    }
                     steps{
                         runPerfTest(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release')
@@ -681,7 +780,7 @@ pipeline {
                     options { retry(2) }
                     agent{ label rocmnode("gfx90a")}
                     environment{
-                        setup_args = "${params.COMPILER_VERSION == "ck-9110" ? """ -DGPU_TARGETS="gfx90a" -DCMAKE_CXX_FLAGS=" -O3 -Xclang -mlink-builtin-bitcode -Xclang /opt/rocm/amdgcn/bitcode/oclc_abi_version_400.bc" -DBUILD_DEV=On """ : """ -DGPU_TARGETS="gfx90a" -DCMAKE_CXX_FLAGS=" -O3 " -DBUILD_DEV=On """}"
+                        setup_args = """ -DGPU_TARGETS="gfx90a" -DBUILD_DEV=On """
                     }
                     steps{
                         runPerfTest(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release')
diff --git a/LICENSE b/LICENSE
index 2fe9a8455efaeda2eab474b2aa038ec2d9e76841..e03fddaf78080705d26ec277629cfb8010c077bf 100644
--- a/LICENSE
+++ b/LICENSE
@@ -7,7 +7,7 @@ Copyright (c) 2020     , Advanced Micro Devices, Inc. (Xiaoyan Zhou)
 Copyright (c) 2021-2022, Advanced Micro Devices, Inc. (Jianfeng Yan)
 
 SPDX-License-Identifier: MIT
-Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
diff --git a/README.md b/README.md
index 151da974a559bef2e52971034f607d4afe44f632..fd6f7e37c2d5cb667cfb20e899b12cc542351101 100644
--- a/README.md
+++ b/README.md
@@ -1,43 +1,60 @@
 # Composable Kernel
 
 ## Methodology
+
 Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
 
 CK utilizes two concepts to achieve performance portability and code maintainability:
 * A tile-based programming model
 * Algorithm complexity reduction for complex ML operators, using innovative technique we call "Tensor Coordinate Transformation".
 
-![ALT](/doc/image/ck_component.png "CK Components")
+![ALT](/docs/data/ck_component.png "CK Components")
 
 ## Code Structure
+
 Current CK library are structured into 4 layers:
 * "Templated Tile Operators" layer
 * "Templated Kernel and Invoker" layer
 * "Instantiated Kernel and Invoker" layer
 * "Client API" layer
 
-![ALT](/doc/image/ck_layer.png "CK Layers")
+![ALT](/docs/data/ck_layer.png "CK Layers")
+
+## Documentation
+
+Run the steps below to build documentation locally.
+
+```
+cd docs
+pip3 install -r sphinx/requirements.txt
+python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
+```
 
 ## Contributors
+
 The list of developers and contributors is here: [Contributors](/CONTRIBUTORS.md)
 
 ## Citation
+
 If you use CK, please use following citations:
 * CK paper will be freely available on arXiv soon: [Realizing Tensor Operators Using Coordinate Transformations and Tile Based Programming](???)
 * [CITATION.cff](/CITATION.cff)
 
 ## License
+
 CK is released under the MIT license. [License File](/LICENSE)
 
 
 # Build CK
 
 ## Build docker image
+
 ```bash
 DOCKER_BUILDKIT=1 docker build -t ck:latest -f Dockerfile .
 ```
 
 ## Launch docker
+
 ```bash
 docker run                                     \
 -it                                            \
@@ -50,10 +67,12 @@ ck:latest                                      \
 ```
 
 ## Build CK
+
 ```bash
 mkdir build && cd build
 
 # Need to specify target ID, example below is for gfx908 and gfx90a
+
 cmake                                                                                             \
 -D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
 -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
@@ -64,6 +83,7 @@ cmake
 ```
 
 ### Build examples and tests
+
 ```bash
  make -j examples tests
  make test
@@ -73,21 +93,43 @@ Instructions for running each individual examples are under [example](/example)
 
 
 ## Build ckProfiler
+
 ```bash
  make -j ckProfiler
 ```
 Instructions for running ckProfiler are under [profiler](/profiler)
 
 ## Install CK
+
 ```bash
 make install
 ```
 
 ## Using CK as pre-built kernel library
+
 Instructions for using CK as a pre-built kernel library are under [client_example](/client_example)
 
+## Contributing
+
+When you contribute to Composable Kernel, make sure to run `clang-format` on all the changed files. We highly recommend using git hooks that are managed by the `pre-commit` framework. To install hooks, run:
+
+```bash
+sudo script/install_precommit.sh
+```
+
+This way, `pre-commit` will add the appropriate hooks to your local repository and automatically run `clang-format` (and possibly additional checks) before any commit is created.
+
+If you need to uninstall hooks from the repository, you can do so by running the following command:
+
+```bash
+script/uninstall_precommit.sh
+```
+
+If for any reason, you need to temporarily disable precommit hooks, you can add the `--no-verify` option to the `git commit` command.
+
 ## Caveat
 ### Kernel Timing and Verification
+
 CK's own kernel timer will warn up kernel once, and then run it multiple times
 to get average kernel time. For some kernels that use atomic add, this will cause
 output buffer to be accumulated multiple times, causing verification failure.
diff --git a/client_example/01_gemm/gemm.cpp b/client_example/01_gemm/gemm.cpp
index a8a6bf16c2b09a3204b119be63258f9ca213336f..c37f208db1cee9bfff1fff469fd79d059fb179f0 100644
--- a/client_example/01_gemm/gemm.cpp
+++ b/client_example/01_gemm/gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
@@ -83,7 +83,7 @@ int main(int argc, char* argv[])
         [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
             using Layout = decltype(layout);
 
-            if(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
             {
                 return (nRow - 1) * stride + nCol;
             }
diff --git a/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt b/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
index b7b724ccc484d6c9ea83e791aa5b8fcde420eef7..ba2952022233b3ffd21fc245b8fd199a5ec5b096 100644
--- a/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
@@ -11,3 +11,17 @@ target_link_libraries(client_gemm_fastgelu PRIVATE composable_kernel::device_ope
 
 add_dependencies(client_gemm_fastgelu_examples client_gemm_add_add_fastgelu client_gemm_add_fastgelu
                  client_gemm_fastgelu)
+
+add_custom_target(client_gemm_fastgelu_generic_examples)
+
+add_executable(client_gemm_add_add_fastgelu_generic gemm_add_add_fastgelu_generic.cpp)
+target_link_libraries(client_gemm_add_add_fastgelu_generic PRIVATE composable_kernel::device_operations)
+
+add_executable(client_gemm_add_fastgelu_generic gemm_add_fastgelu_generic.cpp)
+target_link_libraries(client_gemm_add_fastgelu_generic PRIVATE composable_kernel::device_operations)
+
+add_executable(client_gemm_fastgelu_generic gemm_fastgelu_generic.cpp)
+target_link_libraries(client_gemm_fastgelu_generic PRIVATE composable_kernel::device_operations)
+
+add_dependencies(client_gemm_fastgelu_generic_examples client_gemm_add_add_fastgelu_generic 
+                 client_gemm_add_fastgelu_generic client_gemm_fastgelu_generic)
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
index f88e72b62e4a20beeda8b1d5432a1055c32e2627..756889562e84c66efb5f972621bcb61edda3af82 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
@@ -92,7 +92,7 @@ int main(int argc, char* argv[])
         [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
             using Layout = decltype(layout);
 
-            if(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
             {
                 return (nRow - 1) * stride + nCol;
             }
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu_generic.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu_generic.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..2ed942f0adf64df90ea1046d1de191ce2247d7c4
--- /dev/null
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu_generic.cpp
@@ -0,0 +1,176 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+#include <stdexcept>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough    = ck::tensor_operation::element_wise::PassThrough;
+using AddAddFastGelu = ck::tensor_operation::element_wise::AddAddFastGelu;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddAddFastGelu;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using D0DataType = F16;
+using D1DataType = F16;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using D0Layout = Row;
+using D1Layout = Row;
+using ELayout  = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA  = 4096;
+    ck::index_t StrideB  = 4096;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideD1 = 4096;
+    ck::index_t StrideE  = 4096;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 9)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA  = std::stoi(argv[4]);
+        StrideB  = std::stoi(argv[5]);
+        StrideD0 = std::stoi(argv[6]);
+        StrideD1 = std::stoi(argv[7]);
+        StrideE  = std::stoi(argv[8]);
+    }
+    else
+    {
+        printf("arg1 to 8: M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideE\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem d0_m_n_device_buf(sizeof(D0DataType) *
+                                      f_matrix_space_size(M, N, StrideD0, D0Layout{}));
+    SimpleDeviceMem d1_m_n_device_buf(sizeof(D1DataType) *
+                                      f_matrix_space_size(M, N, StrideD1, D1Layout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<
+        ALayout,
+        BLayout,
+        ck::Tuple<D0Layout, D1Layout>,
+        ELayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<D0DataType, D1DataType>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::AddAddFastGelu>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    // get generic instance
+    auto& op_ptr = op_ptrs[0];
+
+    std::cout << "Run the generic instance without timing: " << op_ptr->GetTypeString()
+              << std::endl;
+
+    // run the generic instance
+    auto argument_ptr =
+        op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 2>{d0_m_n_device_buf.GetDeviceBuffer(),
+                                                               d1_m_n_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    M,
+                                    N,
+                                    K,
+                                    StrideA,
+                                    StrideB,
+                                    std::array<ck::index_t, 2>{StrideD0, StrideD1},
+                                    StrideE,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+    if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+    {
+        invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+    }
+    else
+    {
+        throw std::runtime_error(
+            "Generic instance should be suitable for various input lengths/strides");
+    }
+
+    std::cout << "Done" << std::endl;
+
+    return 0;
+}
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
index 512555f978efc6932e38ff31ec20ebf3aab4b063..8d2a8c234aae63a3566478b5aa9588389a247d4e 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
@@ -76,7 +76,7 @@ int main(int argc, char* argv[])
         StrideA  = std::stoi(argv[4]);
         StrideB  = std::stoi(argv[5]);
         StrideD0 = std::stoi(argv[6]);
-        StrideE  = std::stoi(argv[8]);
+        StrideE  = std::stoi(argv[7]);
     }
     else
     {
@@ -88,7 +88,7 @@ int main(int argc, char* argv[])
         [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
             using Layout = decltype(layout);
 
-            if(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
             {
                 return (nRow - 1) * stride + nCol;
             }
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu_generic.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu_generic.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..644b428fc9f51b28a0f81ed7d79ac741ca6fbcbe
--- /dev/null
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu_generic.cpp
@@ -0,0 +1,169 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+#include <stdexcept>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddFastGelu = ck::tensor_operation::element_wise::AddFastGelu;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddFastGelu;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using D0DataType = F16;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using D0Layout = Row;
+using ELayout  = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA  = 4096;
+    ck::index_t StrideB  = 4096;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideE  = 4096;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 8)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA  = std::stoi(argv[4]);
+        StrideB  = std::stoi(argv[5]);
+        StrideD0 = std::stoi(argv[6]);
+        StrideE  = std::stoi(argv[7]);
+    }
+    else
+    {
+        printf("arg1 to 7: M, N, K, StrideA, StrideB, StrideD0, StrideE\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem d0_m_n_device_buf(sizeof(D0DataType) *
+                                      f_matrix_space_size(M, N, StrideD0, D0Layout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<
+        ALayout,
+        BLayout,
+        ck::Tuple<D0Layout>,
+        ELayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<D0DataType>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::AddFastGelu>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    // get generic instance
+    auto& op_ptr = op_ptrs[0];
+
+    std::cout << "Run the generic instance without timing: " << op_ptr->GetTypeString()
+              << std::endl;
+
+    // run the generic instance
+    auto argument_ptr =
+        op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 1>{d0_m_n_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    M,
+                                    N,
+                                    K,
+                                    StrideA,
+                                    StrideB,
+                                    std::array<ck::index_t, 1>{StrideD0},
+                                    StrideE,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+    if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+    {
+        invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+    }
+    else
+    {
+        throw std::runtime_error(
+            "Generic instance should be suitable for various input lengths/strides");
+    }
+
+    std::cout << "Done" << std::endl;
+
+    return 0;
+}
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
index 72372310321e297b3a0d9101c808eda273675dde..c02df018fd35c6d37a2c6f9b9fded6390c9afb19 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
@@ -72,7 +72,7 @@ int main(int argc, char* argv[])
 
         StrideA = std::stoi(argv[4]);
         StrideB = std::stoi(argv[5]);
-        StrideE = std::stoi(argv[8]);
+        StrideE = std::stoi(argv[6]);
     }
     else
     {
@@ -84,7 +84,7 @@ int main(int argc, char* argv[])
         [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
             using Layout = decltype(layout);
 
-            if(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
             {
                 return (nRow - 1) * stride + nCol;
             }
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu_generic.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu_generic.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..482e93b421f7700cdf37ee014cf407ca3c63555b
--- /dev/null
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu_generic.cpp
@@ -0,0 +1,162 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+#include <stdexcept>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using FastGelu    = ck::tensor_operation::element_wise::FastGelu;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = FastGelu;
+
+using ADataType = F16;
+using BDataType = F16;
+using EDataType = F16;
+
+using ALayout = Row;
+using BLayout = Col;
+using ELayout = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideE = 4096;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 7)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA = std::stoi(argv[4]);
+        StrideB = std::stoi(argv[5]);
+        StrideE = std::stoi(argv[6]);
+    }
+    else
+    {
+        printf("arg1 to 6: M, N, K, StrideA, StrideB, StrideE\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<
+        ALayout,
+        BLayout,
+        ck::Tuple<>,
+        ELayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::FastGelu>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    // get generic instance
+    auto& op_ptr = op_ptrs[0];
+
+    std::cout << "Run the generic instance without timing: " << op_ptr->GetTypeString()
+              << std::endl;
+
+    // run the generic instance
+    auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                                    b_device_buf.GetDeviceBuffer(),
+                                                    {},
+                                                    e_device_buf.GetDeviceBuffer(),
+                                                    M,
+                                                    N,
+                                                    K,
+                                                    StrideA,
+                                                    StrideB,
+                                                    {},
+                                                    StrideE,
+                                                    a_element_op,
+                                                    b_element_op,
+                                                    cde_element_op);
+
+    auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+    if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+    {
+        invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+    }
+    else
+    {
+        throw std::runtime_error(
+            "Generic instance should be suitable for various input lengths/strides");
+    }
+
+    std::cout << "Done" << std::endl;
+
+    return 0;
+}
diff --git a/client_example/03_gemm_layernorm/CMakeLists.txt b/client_example/03_gemm_layernorm/CMakeLists.txt
index 3742e70844b96575e263b22a14b0bb8c4cde7a43..b38698d9064e6fa3cbd2e01c8818c0ebfe361cb3 100644
--- a/client_example/03_gemm_layernorm/CMakeLists.txt
+++ b/client_example/03_gemm_layernorm/CMakeLists.txt
@@ -1,2 +1,5 @@
-add_executable(client_gemm_add_add_reduce_normalize gemm_add_add_layernorm.cpp)
-target_link_libraries(client_gemm_add_add_reduce_normalize PRIVATE composable_kernel::device_operations)
+add_executable(client_gemm_add_add_layernorm_naive gemm_add_add_layernorm_naive.cpp)
+target_link_libraries(client_gemm_add_add_layernorm_naive PRIVATE composable_kernel::device_operations)
+
+add_executable(client_gemm_add_relu_add_layernorm_welford gemm_add_relu_add_layernorm_welford.cpp)
+target_link_libraries(client_gemm_add_relu_add_layernorm_welford PRIVATE composable_kernel::device_operations)
diff --git a/client_example/03_gemm_layernorm/gemm_add_add_layernorm.cpp b/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
similarity index 94%
rename from client_example/03_gemm_layernorm/gemm_add_add_layernorm.cpp
rename to client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
index 6c259407d4608f90bd331ee9a9686b56ad62de90..58c91f903bc7e2f3f296b06c746b1629513bc7f2 100644
--- a/client_example/03_gemm_layernorm/gemm_add_add_layernorm.cpp
+++ b/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
@@ -8,7 +8,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_reduce.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp"
@@ -172,25 +172,26 @@ int main()
             BLayout,
             CLayout>();
 
-    const auto normalize_ptrs =
-        ck::tensor_operation::device::instance::get_device_normalize_from_mean_meansquare_instances<
-            CDataType,
-            ReduceDataType,
-            ReduceDataType,
-            GammaDataType,
-            BetaDataType,
-            LayerNormOutDataType>();
-
     std::cout << "found " << gemm_reduce_ptrs.size()
               << " gemm_reduceMean_reduceSquareMean instances" << std::endl;
 
+    using NormalizeDeviceOp = ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<CDataType, ReduceDataType, ReduceDataType, GammaDataType, BetaDataType>,
+        ck::Tuple<LayerNormOutDataType>,
+        ck::tensor_operation::element_wise::Normalize,
+        2>;
+
+    const auto normalize_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            NormalizeDeviceOp>::GetInstances();
+
     std::cout << "found " << normalize_ptrs.size() << " normalize instances" << std::endl;
 
     auto f_matrix_space_size =
         [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
             using Layout = decltype(layout);
 
-            if(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
             {
                 return (nRow - 1) * stride + nCol;
             }
diff --git a/client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp b/client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..3d5fb6004844af269f7786ae05de8c20cc720633
--- /dev/null
+++ b/client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp
@@ -0,0 +1,244 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using F16 = ck::half_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddReluAdd  = ck::tensor_operation::element_wise::AddReluAdd;
+
+// DataType
+using ADataType     = F16;
+using BDataType     = F16;
+using D0DataType    = F16;
+using D1DataType    = F16;
+using GammaDataType = F16;
+using BetaDataType  = F16;
+using HDataType     = F16;
+
+// Layout
+using ALayout  = Row;
+using BLayout  = Col;
+using D0Layout = Row;
+using D1Layout = Row;
+using HLayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddReluAdd;
+using HElementOp   = PassThrough;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}, mMemSize_(mem_size)
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    void SetZero() const { (void)hipMemset(p_mem_, 0, mMemSize_); }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+    std::size_t mMemSize_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 1024;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
+
+    ck::index_t StrideA  = K;
+    ck::index_t StrideB  = K;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideD1 = N;
+    ck::index_t StrideH  = N;
+
+    float epsilon = 1e-5;
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem d0_device_buf(sizeof(D0DataType) *
+                                  f_matrix_space_size(M, N, StrideD0, D0Layout{}));
+    SimpleDeviceMem d1_device_buf(sizeof(D1DataType) *
+                                  f_matrix_space_size(M, N, StrideD1, D1Layout{}));
+    SimpleDeviceMem gamma_device_buf(sizeof(GammaDataType) * N);
+    SimpleDeviceMem beta_device_buf(sizeof(BetaDataType) * N);
+    SimpleDeviceMem h_device_buf(sizeof(HDataType) * f_matrix_space_size(M, N, StrideH, HLayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleDLayernorm<
+        ALayout,
+        BLayout,
+        ck::Tuple<D0Layout, D1Layout>,
+        HLayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<D0DataType, D1DataType>,
+        GammaDataType,
+        BetaDataType,
+        HDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::AddReluAdd,
+        ck::tensor_operation::element_wise::PassThrough>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+    const auto h_element_op   = HElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b_device_buf.GetDeviceBuffer(),
+            {d0_device_buf.GetDeviceBuffer(), d1_device_buf.GetDeviceBuffer()},
+            gamma_device_buf.GetDeviceBuffer(),
+            beta_device_buf.GetDeviceBuffer(),
+            h_device_buf.GetDeviceBuffer(),
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            {StrideD0, StrideD1},
+            StrideH,
+            epsilon,
+            a_element_op,
+            b_element_op,
+            cde_element_op,
+            h_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+            SimpleDeviceMem workspace_dev(workspace_sz);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+            h_device_buf.SetZero();
+
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_byte =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                (sizeof(D0DataType) + sizeof(D1DataType) + sizeof(HDataType)) * M * N +
+                (sizeof(GammaDataType) + sizeof(BetaDataType)) * N;
+
+            float gb_per_sec = num_byte / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+              << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b_device_buf.GetDeviceBuffer(),
+            {d0_device_buf.GetDeviceBuffer(), d1_device_buf.GetDeviceBuffer()},
+            gamma_device_buf.GetDeviceBuffer(),
+            beta_device_buf.GetDeviceBuffer(),
+            h_device_buf.GetDeviceBuffer(),
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            {StrideD0, StrideD1},
+            StrideH,
+            epsilon,
+            a_element_op,
+            b_element_op,
+            cde_element_op,
+            h_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+            SimpleDeviceMem workspace_dev(workspace_sz);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+            h_device_buf.SetZero();
+
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
\ No newline at end of file
diff --git a/client_example/04_contraction/CMakeLists.txt b/client_example/04_contraction/CMakeLists.txt
index 4bc6780f96d2fe4a4912e3c188b4b5155cc162dd..7ffedfeef36839c1c8bb942ffae11e5f6dcadf22 100644
--- a/client_example/04_contraction/CMakeLists.txt
+++ b/client_example/04_contraction/CMakeLists.txt
@@ -1,6 +1,15 @@
-add_executable(client_contraction_scale contraction_scale.cpp)
-target_link_libraries(client_contraction_scale PRIVATE composable_kernel::device_operations)
+add_executable(client_contraction_scale_fp32 contraction_scale_fp32.cpp)
+target_link_libraries(client_contraction_scale_fp32 PRIVATE composable_kernel::device_operations)
 
-add_executable(client_contraction_bilinear contraction_bilinear.cpp)
-target_link_libraries(client_contraction_bilinear PRIVATE composable_kernel::device_operations)
+add_executable(client_contraction_bilinear_fp32 contraction_bilinear_fp32.cpp)
+target_link_libraries(client_contraction_bilinear_fp32 PRIVATE composable_kernel::device_operations)
+
+add_executable(client_contraction_scale_fp64 contraction_scale_fp64.cpp)
+target_link_libraries(client_contraction_scale_fp64 PRIVATE composable_kernel::device_operations)
+
+add_executable(client_contraction_bilinear_fp64 contraction_bilinear_fp64.cpp)
+target_link_libraries(client_contraction_bilinear_fp64 PRIVATE composable_kernel::device_operations)
+
+add_executable(contraction_g1m2n3k1_add_xdl_fp16 contraction_g1m2n3k1_add_xdl_fp16.cpp)
+target_link_libraries(contraction_g1m2n3k1_add_xdl_fp16 PRIVATE composable_kernel::device_operations)
 
diff --git a/client_example/04_contraction/contraction_bilinear.cpp b/client_example/04_contraction/contraction_bilinear_fp32.cpp
similarity index 99%
rename from client_example/04_contraction/contraction_bilinear.cpp
rename to client_example/04_contraction/contraction_bilinear_fp32.cpp
index 91dead41a4cac19db857b99a233839e9e6647c57..89f834b9824e134f8f0aeed8aa54f78a5c8824a3 100644
--- a/client_example/04_contraction/contraction_bilinear.cpp
+++ b/client_example/04_contraction/contraction_bilinear_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <numeric>
diff --git a/client_example/04_contraction/contraction_bilinear_fp64.cpp b/client_example/04_contraction/contraction_bilinear_fp64.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1aa3ba7de597a0b97a295b0f7ee7ad21b1e9cd80
--- /dev/null
+++ b/client_example/04_contraction/contraction_bilinear_fp64.cpp
@@ -0,0 +1,281 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <numeric>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp"
+#include "ck/library/utility/numeric.hpp"
+
+using F64 = double;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = Bilinear;
+
+using ADataType        = F64;
+using BDataType        = F64;
+using AccDataType      = F64;
+using CShuffleDataType = F64;
+using DDataType        = F64;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F64;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+// kknn
+#if 1
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+// knnn
+#elif 0
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{64, 1, 131072, 2048};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+// mknn
+#elif 0
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{128, 1, 245760, 3840};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+// mnnn
+#elif 0
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{128, 1, 245760, 3840};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{64, 1, 131072, 2048};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+#endif
+
+    float alpha = 1.f;
+    float beta  = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 25)
+    {
+        const ck::index_t M0 = std::stoi(argv[1]);
+        const ck::index_t M1 = std::stoi(argv[2]);
+
+        const ck::index_t N0 = std::stoi(argv[3]);
+        const ck::index_t N1 = std::stoi(argv[4]);
+
+        const ck::index_t K0 = std::stoi(argv[5]);
+        const ck::index_t K1 = std::stoi(argv[6]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[7]), std::stoi(argv[8]), std::stoi(argv[9]), std::stoi(argv[10])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13]), std::stoi(argv[14])};
+
+        d_ms_ns_lengths = {M0, M1, N0, N1};
+        d_ms_ns_strides = {
+            std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17]), std::stoi(argv[18])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21]), std::stoi(argv[22])};
+
+        alpha = std::stof(argv[23]);
+        beta  = std::stof(argv[24]);
+    }
+    else
+    {
+        printf("arg1 to 6: M0, M1, N0, N1, K0, K1\n");
+        printf("arg7 to 10: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg11 to 14: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg15 to 18: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
+        printf("arg19 to 22: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg23 to 24: alpha, beta\n");
+        exit(0);
+    }
+
+    auto f_tensor_space_size = [](auto lengths, auto strides) {
+        std::size_t space_size = 1;
+        for(std::size_t i = 0; i < lengths.size(); ++i)
+        {
+            space_size += (lengths[i] - 1) * strides[i];
+        }
+        return space_size;
+    };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) *
+                                 f_tensor_space_size(a_ms_ks_lengths, a_ms_ks_strides));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) *
+                                 f_tensor_space_size(b_ns_ks_lengths, b_ns_ks_strides));
+    SimpleDeviceMem d_device_buf(sizeof(DDataType) *
+                                 f_tensor_space_size(d_ms_ns_lengths, d_ms_ns_strides));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) *
+                                 f_tensor_space_size(e_ms_ns_lengths, e_ms_ns_strides));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceContractionMultipleD<
+        NumDimM,
+        NumDimN,
+        NumDimK,
+        ADataType,
+        BDataType,
+        ck::Tuple<DDataType>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::Bilinear>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{alpha, beta};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                        b_device_buf.GetDeviceBuffer(),
+                                        std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                                        e_device_buf.GetDeviceBuffer(),
+                                        a_ms_ks_lengths,
+                                        a_ms_ks_strides,
+                                        b_ns_ks_lengths,
+                                        b_ns_ks_strides,
+                                        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
+                                        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
+                                        e_ms_ns_lengths,
+                                        e_ms_ns_strides,
+                                        a_element_op,
+                                        b_element_op,
+                                        cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            ck::index_t M = ck::accumulate_n<ck::index_t>(
+                e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+            ck::index_t N = ck::accumulate_n<ck::index_t>(
+                e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+            ck::index_t K = ck::accumulate_n<ck::index_t>(
+                a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+            std::size_t flop      = std::size_t(2) * M * N * K;
+            std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                                    sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return 0;
+}
diff --git a/client_example/04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp b/client_example/04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f8ea2258c2ba262e9db42f1b4dc92ff16cdc6286
--- /dev/null
+++ b/client_example/04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp
@@ -0,0 +1,204 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <numeric>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp"
+#include "ck/library/utility/numeric.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = Add;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F16;
+using DDataType        = F16;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F16;
+
+static constexpr ck::index_t NumDimG = 1;
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 3;
+static constexpr ck::index_t NumDimK = 1;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t G0 = 1;
+
+    ck::index_t M0 = 64;
+    ck::index_t M1 = 256;
+
+    ck::index_t N0 = 3;
+    ck::index_t N1 = 12;
+    ck::index_t N2 = 64;
+
+    ck::index_t K0 = 768;
+
+    // A[M0, M1, M2, K0]
+    std::vector<ck::index_t> a_gs_ms_ks_lengths{G0, M0, M1, K0};
+    std::vector<ck::index_t> a_gs_ms_ks_strides{M0 * M1 * K0, M1 * K0, K0, 1};
+    // B[N0, N1, N2, K0]
+    std::vector<ck::index_t> b_gs_ns_ks_lengths{G0, N0, N1, N2, K0};
+    std::vector<ck::index_t> b_gs_ns_ks_strides{N0 * N1 * N2 * K0, N1 * N2 * K0, N2 * K0, K0, 1};
+
+    // D[N0, M0, N1, M1, N2]
+    std::vector<ck::index_t> d_gs_ms_ns_lengths{G0, M0, M1, N0, N1, N2};
+    std::vector<ck::index_t> d_gs_ms_ns_strides{N0 * N1 * N2, 0, 0, N1 * N2, N2, 1};
+    // E[N0 M0 N1 N2 M1]
+    std::vector<ck::index_t> e_gs_ms_ns_lengths{G0, M0, M1, N0, N1, N2};
+    std::vector<ck::index_t> e_gs_ms_ns_strides{
+        M0 * M1 * N0 * N1 * N2, N1 * N2 * M1, 1, M0 * N1 * N2 * M1, M1 * N2, M1};
+
+    auto f_tensor_space_size = [](auto lengths, auto strides) {
+        std::size_t space_size = 1;
+        for(std::size_t i = 0; i < lengths.size(); ++i)
+        {
+            space_size += (lengths[i] - 1) * strides[i];
+        }
+        return space_size;
+    };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) *
+                                 f_tensor_space_size(a_gs_ms_ks_lengths, a_gs_ms_ks_strides));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) *
+                                 f_tensor_space_size(b_gs_ns_ks_lengths, b_gs_ns_ks_strides));
+    SimpleDeviceMem d_device_buf(sizeof(DDataType) *
+                                 f_tensor_space_size(d_gs_ms_ns_lengths, d_gs_ms_ns_strides));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) *
+                                 f_tensor_space_size(e_gs_ms_ns_lengths, e_gs_ms_ns_strides));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceBatchedContractionMultipleD<
+        NumDimG,
+        NumDimM,
+        NumDimN,
+        NumDimK,
+        ADataType,
+        BDataType,
+        DsDataType,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::Add>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                        b_device_buf.GetDeviceBuffer(),
+                                        std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                                        e_device_buf.GetDeviceBuffer(),
+                                        a_gs_ms_ks_lengths,
+                                        a_gs_ms_ks_strides,
+                                        b_gs_ns_ks_lengths,
+                                        b_gs_ns_ks_strides,
+                                        std::array<std::vector<ck::index_t>, 1>{d_gs_ms_ns_lengths},
+                                        std::array<std::vector<ck::index_t>, 1>{d_gs_ms_ns_strides},
+                                        e_gs_ms_ns_lengths,
+                                        e_gs_ms_ns_strides,
+                                        a_element_op,
+                                        b_element_op,
+                                        cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            ck::index_t M = ck::accumulate_n<ck::index_t>(
+                e_gs_ms_ns_lengths.begin() + NumDimG, NumDimM, 1, std::multiplies<>{});
+
+            ck::index_t N = ck::accumulate_n<ck::index_t>(
+                e_gs_ms_ns_lengths.begin() + NumDimG + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+            ck::index_t K = ck::accumulate_n<ck::index_t>(
+                a_gs_ms_ks_lengths.begin() + NumDimG + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+            std::size_t flop      = std::size_t(2) * M * N * K;
+            std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                                    sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return 0;
+}
diff --git a/client_example/04_contraction/contraction_scale.cpp b/client_example/04_contraction/contraction_scale_fp32.cpp
similarity index 99%
rename from client_example/04_contraction/contraction_scale.cpp
rename to client_example/04_contraction/contraction_scale_fp32.cpp
index 4e08ee19cdb098b2dfb70a662d59c87008400123..ba7b0633c33aabeeb06547edfb08f506e637e599 100644
--- a/client_example/04_contraction/contraction_scale.cpp
+++ b/client_example/04_contraction/contraction_scale_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <numeric>
diff --git a/client_example/04_contraction/contraction_scale_fp64.cpp b/client_example/04_contraction/contraction_scale_fp64.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..24e52eb5aa423339ff96ad0914dc479d715fe7b7
--- /dev/null
+++ b/client_example/04_contraction/contraction_scale_fp64.cpp
@@ -0,0 +1,270 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <numeric>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/contraction_scale.hpp"
+#include "ck/library/utility/numeric.hpp"
+
+using F64 = double;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = Scale;
+
+using ADataType        = F64;
+using BDataType        = F64;
+using AccDataType      = F64;
+using CShuffleDataType = F64;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F64;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+// kkn
+#if 1
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+// knn
+#elif 0
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{64, 1, 131072, 2048};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+// mkn
+#elif 0
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{128, 1, 245760, 3840};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+// mnn
+#elif 0
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{128, 1, 245760, 3840};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{64, 1, 131072, 2048};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+#endif
+
+    float scale = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 20)
+    {
+        const ck::index_t M0 = std::stoi(argv[1]);
+        const ck::index_t M1 = std::stoi(argv[2]);
+
+        const ck::index_t N0 = std::stoi(argv[3]);
+        const ck::index_t N1 = std::stoi(argv[4]);
+
+        const ck::index_t K0 = std::stoi(argv[5]);
+        const ck::index_t K1 = std::stoi(argv[6]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[7]), std::stoi(argv[8]), std::stoi(argv[9]), std::stoi(argv[10])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13]), std::stoi(argv[14])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17]), std::stoi(argv[18])};
+
+        scale = std::stof(argv[19]);
+    }
+    else
+    {
+        printf("arg1 to 6: M0, M1, N0, N1, K0, K1\n");
+        printf("arg7 to 10: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg11 to 14: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg15 to 18: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg19: scale\n");
+        exit(0);
+    }
+
+    auto f_tensor_space_size = [](auto lengths, auto strides) {
+        std::size_t space_size = 1;
+        for(std::size_t i = 0; i < lengths.size(); ++i)
+        {
+            space_size += (lengths[i] - 1) * strides[i];
+        }
+        return space_size;
+    };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) *
+                                 f_tensor_space_size(a_ms_ks_lengths, a_ms_ks_strides));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) *
+                                 f_tensor_space_size(b_ns_ks_lengths, b_ns_ks_strides));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) *
+                                 f_tensor_space_size(e_ms_ns_lengths, e_ms_ns_strides));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceContractionMultipleD<
+        NumDimM,
+        NumDimN,
+        NumDimK,
+        ADataType,
+        BDataType,
+        ck::Tuple<>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::Scale>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{scale};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                                        b_device_buf.GetDeviceBuffer(),
+                                                        std::array<const void*, 0>{},
+                                                        e_device_buf.GetDeviceBuffer(),
+                                                        a_ms_ks_lengths,
+                                                        a_ms_ks_strides,
+                                                        b_ns_ks_lengths,
+                                                        b_ns_ks_strides,
+                                                        std::array<std::vector<ck::index_t>, 0>{},
+                                                        std::array<std::vector<ck::index_t>, 0>{},
+                                                        e_ms_ns_lengths,
+                                                        e_ms_ns_strides,
+                                                        a_element_op,
+                                                        b_element_op,
+                                                        cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            ck::index_t M = ck::accumulate_n<ck::index_t>(
+                e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+            ck::index_t N = ck::accumulate_n<ck::index_t>(
+                e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+            ck::index_t K = ck::accumulate_n<ck::index_t>(
+                a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+            std::size_t flop = std::size_t(2) * M * N * K;
+            std::size_t num_btype =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return 0;
+}
diff --git a/client_example/05_layernorm/layernorm2d.cpp b/client_example/05_layernorm/layernorm2d.cpp
index adb41171e12a87ffafe42e4f112a3e89cfc7296e..4af4d7abe8ee6c5120f6060c78141021052cb619 100644
--- a/client_example/05_layernorm/layernorm2d.cpp
+++ b/client_example/05_layernorm/layernorm2d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
@@ -12,12 +12,12 @@
 
 #include "ck/library/tensor_operation_instance/gpu/normalization.hpp"
 
-using XDataType     = ck::half_t;
-using GammaDataType = ck::half_t;
-using BetaDataType  = ck::half_t;
-using YDataType     = ck::half_t;
-using AccDataType   = float;
-using PassThrough   = ck::tensor_operation::element_wise::PassThrough;
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
 
 constexpr int Rank         = 2;
 constexpr int NumReduceDim = 1;
@@ -54,7 +54,7 @@ int main(int argc, char* argv[])
     using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
                                                                        GammaDataType,
                                                                        BetaDataType,
-                                                                       AccDataType,
+                                                                       ComputeDataType,
                                                                        YDataType,
                                                                        PassThrough,
                                                                        Rank,
diff --git a/client_example/06_softmax/softmax4d.cpp b/client_example/06_softmax/softmax4d.cpp
index 7745ddf34cf7abb65265ada74f0896dedcbbf655..2ccad27a88757b576a050d0558acc4108857cbd1 100644
--- a/client_example/06_softmax/softmax4d.cpp
+++ b/client_example/06_softmax/softmax4d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
@@ -47,18 +47,41 @@ int main(int argc, char* argv[])
     ck::index_t num_elements =
         std::accumulate(in_lengths.begin(), in_lengths.end(), 1, std::multiplies<ck::index_t>());
 
-    AccDataType alpha{2.0f};
-    AccDataType beta{2.0f};
+    double alpha{2.0};
+    double beta{2.0};
 
     SimpleDeviceMem in(sizeof(InDataType) * num_elements);
     SimpleDeviceMem out(sizeof(OutDataType) * num_elements);
 
-    using DeviceOp = ck::tensor_operation::device::
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>;
+    using DeviceOp = ck::tensor_operation::device::DeviceSoftmax<InDataType,
+                                                                 AccDataType,
+                                                                 OutDataType,
+                                                                 PassThrough,
+                                                                 PassThrough,
+                                                                 Rank,
+                                                                 NumReduceDim>;
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
         DeviceOp>::GetInstances();
 
+    auto& generic_op_ptr = op_ptrs[0];
+
+    auto generic_argument_ptr = generic_op_ptr->MakeArgumentPointer(in_lengths,
+                                                                    in_strides,
+                                                                    reduce_dims,
+                                                                    alpha,
+                                                                    beta,
+                                                                    in.GetDeviceBuffer(),
+                                                                    out.GetDeviceBuffer(),
+                                                                    PassThrough{},
+                                                                    PassThrough{});
+
+    if(!generic_op_ptr->IsSupportedArgument(generic_argument_ptr.get()))
+    {
+        throw std::runtime_error(
+            "The generic kernel instance should be able to support any input shapes");
+    };
+
     std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
 
     std::string best_op_name;
@@ -74,16 +97,11 @@ int main(int argc, char* argv[])
     {
         auto& op_ptr = op_ptrs[i];
 
-        if(op_ptr->GetRank() != Rank || op_ptr->GetNumReduceDim() != NumReduceDim)
-        {
-            continue;
-        }
-
         auto argument_ptr   = op_ptr->MakeArgumentPointer(in_lengths,
                                                         in_strides,
                                                         reduce_dims,
-                                                        &alpha,
-                                                        &beta,
+                                                        alpha,
+                                                        beta,
                                                         in.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
                                                         PassThrough{},
@@ -129,8 +147,8 @@ int main(int argc, char* argv[])
         auto argument_ptr = op_ptr->MakeArgumentPointer(in_lengths,
                                                         in_strides,
                                                         reduce_dims,
-                                                        &alpha,
-                                                        &beta,
+                                                        alpha,
+                                                        beta,
                                                         in.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
                                                         PassThrough{},
@@ -147,4 +165,4 @@ int main(int argc, char* argv[])
     }
 
     return 0;
-}
\ No newline at end of file
+}
diff --git a/client_example/07_grouped_conv2d_fwd/CMakeLists.txt b/client_example/07_grouped_conv2d_fwd/CMakeLists.txt
deleted file mode 100644
index ddc83168acfbbdf3d58b0909b761947c792a3c06..0000000000000000000000000000000000000000
--- a/client_example/07_grouped_conv2d_fwd/CMakeLists.txt
+++ /dev/null
@@ -1,2 +0,0 @@
-add_executable(client_grouped_conv2d_fwd grouped_conv2d_fwd.cpp)
-target_link_libraries(client_grouped_conv2d_fwd PRIVATE composable_kernel::device_operations)
diff --git a/client_example/07_grouped_convnd_fwd/CMakeLists.txt b/client_example/07_grouped_convnd_fwd/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fce7e91c1ef5b23fcfecfe99cc49263a1d779d76
--- /dev/null
+++ b/client_example/07_grouped_convnd_fwd/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_executable(client_grouped_conv2d_fwd grouped_conv2d_fwd.cpp)
+target_link_libraries(client_grouped_conv2d_fwd PRIVATE composable_kernel::device_operations)
+
+add_executable(client_grouped_conv1d_fwd grouped_conv1d_fwd.cpp)
+target_link_libraries(client_grouped_conv1d_fwd PRIVATE composable_kernel::device_operations)
diff --git a/client_example/07_grouped_conv2d_fwd/grouped_conv2d_fwd.cpp b/client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp
similarity index 82%
rename from client_example/07_grouped_conv2d_fwd/grouped_conv2d_fwd.cpp
rename to client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp
index ece6e30c56004f58ff136502b4d8874e8269cd3a..70be0101c6d92512ab28a72797d2b8c46fb55281 100644
--- a/client_example/07_grouped_conv2d_fwd/grouped_conv2d_fwd.cpp
+++ b/client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iomanip>
@@ -17,21 +17,18 @@ using InDataType  = ck::half_t;
 using WeiDataType = ck::half_t;
 using OutDataType = ck::half_t;
 
-using InLayout    = ck::tensor_layout::convolution::GNHWC;
-using WeiLayout   = ck::tensor_layout::convolution::GKYXC;
-using OutLayout   = ck::tensor_layout::convolution::GNHWK;
+using InLayout    = ck::tensor_layout::convolution::GNWC;
+using WeiLayout   = ck::tensor_layout::convolution::GKXC;
+using OutLayout   = ck::tensor_layout::convolution::GNWK;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t NumDimSpatial = 1;
 static constexpr ck::index_t G             = 32;
 static constexpr ck::index_t N             = 256;
 static constexpr ck::index_t K             = 192;
 static constexpr ck::index_t C             = 192;
-static constexpr ck::index_t Y             = 3;
 static constexpr ck::index_t X             = 3;
-static constexpr ck::index_t Hi            = 28;
 static constexpr ck::index_t Wi            = 28;
-static constexpr ck::index_t Ho            = 28;
 static constexpr ck::index_t Wo            = 28;
 
 struct SimpleDeviceMem
@@ -52,14 +49,14 @@ struct SimpleDeviceMem
 
 int main()
 {
-    std::array<ck::index_t, NumDimSpatial + 3> in_lengths{G, N, Hi, Wi, C};
-    std::array<ck::index_t, NumDimSpatial + 3> in_strides{0, 0, 0, 0, 1};
+    std::array<ck::index_t, NumDimSpatial + 3> in_lengths{G, N, Wi, C};
+    std::array<ck::index_t, NumDimSpatial + 3> in_strides{0, 0, 0, 1};
 
-    std::array<ck::index_t, NumDimSpatial + 3> wei_lengths{G, K, Y, X, C};
-    std::array<ck::index_t, NumDimSpatial + 3> wei_strides{0, 0, 0, 0, 1};
+    std::array<ck::index_t, NumDimSpatial + 3> wei_lengths{G, K, X, C};
+    std::array<ck::index_t, NumDimSpatial + 3> wei_strides{0, 0, 0, 1};
 
-    std::array<ck::index_t, NumDimSpatial + 3> out_lengths{G, N, Ho, Wo, K};
-    std::array<ck::index_t, NumDimSpatial + 3> out_strides{0, 0, 0, 0, 1};
+    std::array<ck::index_t, NumDimSpatial + 3> out_lengths{G, N, Wo, K};
+    std::array<ck::index_t, NumDimSpatial + 3> out_strides{0, 0, 0, 1};
 
     std::partial_sum(rbegin(in_lengths),
                      std::prev(rend(in_lengths)),
@@ -74,28 +71,34 @@ int main()
                      std::next(rbegin(out_strides)),
                      std::multiplies<>{});
 
-    // transpose GNHWC/GKYXC/GNHWK to GNCHW/GKCYX/GNCHW
-    std::rotate(
-        rbegin(in_lengths), std::next(rbegin(in_lengths)), std::next(rbegin(in_lengths), 3));
-    std::rotate(
-        rbegin(in_strides), std::next(rbegin(in_strides)), std::next(rbegin(in_strides), 3));
-    std::rotate(
-        rbegin(wei_lengths), std::next(rbegin(wei_lengths)), std::next(rbegin(wei_lengths), 3));
-    std::rotate(
-        rbegin(wei_strides), std::next(rbegin(wei_strides)), std::next(rbegin(wei_strides), 3));
-    std::rotate(
-        rbegin(out_lengths), std::next(rbegin(out_lengths)), std::next(rbegin(out_lengths), 3));
-    std::rotate(
-        rbegin(out_strides), std::next(rbegin(out_strides)), std::next(rbegin(out_strides), 3));
-
-    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
-    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
-    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
-    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
-
-    SimpleDeviceMem in(sizeof(InDataType) * G * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Ho * Wo * K);
+    // transpose GNWC/GKXC/GNWK to GNCW/GKCX/GNCW
+    std::rotate(rbegin(in_lengths),
+                std::next(rbegin(in_lengths)),
+                std::next(rbegin(in_lengths), NumDimSpatial + 1));
+    std::rotate(rbegin(in_strides),
+                std::next(rbegin(in_strides)),
+                std::next(rbegin(in_strides), NumDimSpatial + 1));
+    std::rotate(rbegin(wei_lengths),
+                std::next(rbegin(wei_lengths)),
+                std::next(rbegin(wei_lengths), NumDimSpatial + 1));
+    std::rotate(rbegin(wei_strides),
+                std::next(rbegin(wei_strides)),
+                std::next(rbegin(wei_strides), NumDimSpatial + 1));
+    std::rotate(rbegin(out_lengths),
+                std::next(rbegin(out_lengths)),
+                std::next(rbegin(out_lengths), NumDimSpatial + 1));
+    std::rotate(rbegin(out_strides),
+                std::next(rbegin(out_strides)),
+                std::next(rbegin(out_strides), NumDimSpatial + 1));
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * G * N * Wi * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Wo * K);
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
                                                                                  InLayout,
@@ -154,10 +157,10 @@ int main()
         {
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
-            std::size_t flop      = std::size_t(2) * G * N * K * C * Ho * Wo * Y * X;
-            std::size_t num_bytes = sizeof(InDataType) * G * N * Hi * Wi * C +
-                                    sizeof(WeiDataType) * G * K * Y * X * C +
-                                    sizeof(OutDataType) * G * N * Ho * Wo * K;
+            std::size_t flop      = std::size_t(2) * G * N * K * C * Wo * X;
+            std::size_t num_bytes = sizeof(InDataType) * G * N * Wi * C +
+                                    sizeof(WeiDataType) * G * K * X * C +
+                                    sizeof(OutDataType) * G * N * Wo * K;
 
             float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight.cpp b/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
similarity index 60%
rename from client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight.cpp
rename to client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
index 1ecc8568959555740e88892ff6233ca1a5cf7333..57a210fa1f5f0d3f0eca014e08ccd119d29e4fd6 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight.cpp
+++ b/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
@@ -1,38 +1,38 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iomanip>
 #include <iostream>
 #include <iterator>
 #include <numeric>
+#include <vector>
 
 #include "ck/ck.hpp"
-#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 using InDataType  = ck::half_t;
 using WeiDataType = ck::half_t;
 using OutDataType = ck::half_t;
 
-using InLayout    = ck::tensor_layout::convolution::GNHWC;
+using InLayout    = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout   = ck::tensor_layout::convolution::GKYXC;
-using OutLayout   = ck::tensor_layout::convolution::GNHWK;
+using OutLayout   = ck::tensor_layout::convolution::NHWGK;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr ck::index_t NumDimSpatial = 2;
 static constexpr ck::index_t G             = 32;
-static constexpr ck::index_t N             = 256;
-static constexpr ck::index_t K             = 192;
-static constexpr ck::index_t C             = 192;
-static constexpr ck::index_t Y             = 3;
-static constexpr ck::index_t X             = 3;
-static constexpr ck::index_t Hi            = 28;
-static constexpr ck::index_t Wi            = 28;
-static constexpr ck::index_t Ho            = 28;
-static constexpr ck::index_t Wo            = 28;
+static constexpr ck::index_t N             = 256; // batch size
+static constexpr ck::index_t K             = 64;  // output channel
+static constexpr ck::index_t C             = 32;  // input channel (per group)
+static constexpr ck::index_t Y             = 3;   // filter H
+static constexpr ck::index_t X             = 3;   // filter W
+static constexpr ck::index_t Hi            = 28;  // input H
+static constexpr ck::index_t Wi            = 28;  // input W
+static constexpr ck::index_t Ho            = 28;  // output H
+static constexpr ck::index_t Wo            = 28;  // output W
 
 struct SimpleDeviceMem
 {
@@ -52,31 +52,38 @@ struct SimpleDeviceMem
 
 int main()
 {
-    std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Hi, Wi};
-    std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Y, X};
-    std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Ho, Wo};
-
-    std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1};
-    std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1};
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
+    std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
+    std::array<ck::index_t, 5> wei_lengths{G, K, C, Y, X};
+    std::array<ck::index_t, 5> wei_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
     std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
     std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
 
-    ck::index_t split_k = 2;
-
-    SimpleDeviceMem in(sizeof(InDataType) * G * N * Hi * Wi * C);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
     SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Ho * Wo * K);
-
-    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdWeight<NumDimSpatial,
-                                                                              InLayout,
-                                                                              WeiLayout,
-                                                                              OutLayout,
-                                                                              InDataType,
-                                                                              WeiDataType,
-                                                                              OutDataType,
-                                                                              PassThrough,
-                                                                              PassThrough,
-                                                                              PassThrough>;
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
+                                                                                 InLayout,
+                                                                                 WeiLayout,
+                                                                                 ck::Tuple<>,
+                                                                                 OutLayout,
+                                                                                 InDataType,
+                                                                                 WeiDataType,
+                                                                                 ck::Tuple<>,
+                                                                                 OutDataType,
+                                                                                 PassThrough,
+                                                                                 PassThrough,
+                                                                                 PassThrough>;
+
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
         DeviceOp>::GetInstances();
@@ -97,22 +104,23 @@ int main()
         auto& op_ptr        = op_ptrs[i];
         auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         wei.GetDeviceBuffer(),
+                                                        {},
                                                         out.GetDeviceBuffer(),
-                                                        G,
-                                                        N,
-                                                        K,
-                                                        C,
-                                                        input_spatial_lengths,
-                                                        filter_spatial_lengths,
-                                                        output_spatial_lengths,
-                                                        conv_filter_strides,
-                                                        conv_filter_dilations,
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
                                                         input_left_pads,
                                                         input_right_pads,
                                                         PassThrough{},
                                                         PassThrough{},
-                                                        PassThrough{},
-                                                        split_k);
+                                                        PassThrough{});
         auto invoker_ptr    = op_ptr->MakeInvokerPointer();
         std::string op_name = op_ptr->GetTypeString();
 
@@ -121,9 +129,9 @@ int main()
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
             std::size_t flop      = std::size_t(2) * G * N * K * C * Ho * Wo * Y * X;
-            std::size_t num_bytes = sizeof(InDataType) * G * N * Hi * Wi * C +
+            std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
                                     sizeof(WeiDataType) * G * K * Y * X * C +
-                                    sizeof(OutDataType) * G * N * Ho * Wo * K;
+                                    sizeof(OutDataType) * N * Ho * Wo * G * K;
 
             float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
@@ -162,23 +170,25 @@ int main()
                   << std::endl;
         auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         wei.GetDeviceBuffer(),
+                                                        {},
                                                         out.GetDeviceBuffer(),
-                                                        G,
-                                                        N,
-                                                        K,
-                                                        C,
-                                                        input_spatial_lengths,
-                                                        filter_spatial_lengths,
-                                                        output_spatial_lengths,
-                                                        conv_filter_strides,
-                                                        conv_filter_dilations,
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
                                                         input_left_pads,
                                                         input_right_pads,
                                                         PassThrough{},
                                                         PassThrough{},
-                                                        PassThrough{},
-                                                        split_k);
-        auto invoker_ptr  = op_ptr->MakeInvokerPointer();
+                                                        PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
         if(op_ptr->IsSupportedArgument(argument_ptr.get()))
         {
diff --git a/client_example/08_fused_attention/CMakeLists.txt b/client_example/08_fused_attention/CMakeLists.txt
index 5cdea72fd996fefbc8eac3781199ada4e3ce1fb0..862b9ed5b70ee8a35c4ba44e2ccf1441e1532bd3 100644
--- a/client_example/08_fused_attention/CMakeLists.txt
+++ b/client_example/08_fused_attention/CMakeLists.txt
@@ -1,2 +1,5 @@
 add_executable(client_fused_attention fused_attention.cpp)
 target_link_libraries(client_fused_attention PRIVATE composable_kernel::device_operations)
+
+add_executable(client_fused_attention_bias fused_attention_bias.cpp)
+target_link_libraries(client_fused_attention_bias PRIVATE composable_kernel::device_operations)
diff --git a/client_example/08_fused_attention/fused_attention.cpp b/client_example/08_fused_attention/fused_attention.cpp
index fe927da1248786a4b943f610ce38b75f0d88defd..df6bc11a70d32df221612466a8af0fbcd9cafb1c 100644
--- a/client_example/08_fused_attention/fused_attention.cpp
+++ b/client_example/08_fused_attention/fused_attention.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/client_example/08_fused_attention/fused_attention_bias.cpp b/client_example/08_fused_attention/fused_attention_bias.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..6c9f3bc8f6f5a3c06f339f1246b5b3985e11d2d8
--- /dev/null
+++ b/client_example/08_fused_attention/fused_attention_bias.cpp
@@ -0,0 +1,226 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using AElementOp    = ck::tensor_operation::element_wise::PassThrough;
+using B0ElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using Acc0ElementOp = ck::tensor_operation::element_wise::ScaleAdd;
+using B1ElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CElementOp    = ck::tensor_operation::element_wise::PassThrough;
+
+constexpr static auto MaskingSpec =
+    ck::tensor_operation::device::MaskingSpecialization::MaskDisabled;
+
+using ADataType   = ck::half_t;
+using B0DataType  = ck::half_t;
+using B1DataType  = ck::half_t;
+using CDataType   = ck::half_t;
+using D0DataType  = ck::half_t;
+using AccDataType = float;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    int G0 = 48;
+    int G1 = 16;
+    int M  = 1024;
+    int N  = 1024;
+    int K  = 64;
+    int O  = 64;
+
+    // A layout [G0, M, G1, K]
+    std::vector<ck::index_t> a_gs_ms_ks_lengths{G0, G1, M, K};
+    std::vector<ck::index_t> a_gs_ms_ks_strides{M * G1 * K, K, G1 * K, 1};
+
+    // B0 layout [G0, N, G1, K]
+    std::vector<ck::index_t> b0_gs_ns_ks_lengths{G0, G1, N, K};
+    std::vector<ck::index_t> b0_gs_ns_ks_strides{N * G1 * K, K, G1 * K, 1};
+
+    // B1 layout [G0, N, G1, O]
+    std::vector<ck::index_t> b1_gs_os_ns_lengths{G0, G1, O, N};
+    std::vector<ck::index_t> b1_gs_os_ns_strides{N * G1 * O, O, 1, G1 * O};
+
+    // C layout [G0, M, G1, O]
+    std::vector<ck::index_t> c_gs_ms_os_lengths{G0, G1, M, O};
+    std::vector<ck::index_t> c_gs_ms_os_strides{M * G1 * O, O, G1 * O, 1};
+
+    // D layout [G0, M, G1, N]
+    std::vector<ck::index_t> d0_gs_ms_ns_lengths{G0, G1, M, N};
+    std::vector<ck::index_t> d0_gs_ms_ns_strides{M * G1 * N, N, G1 * N, 1};
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * G0 * G1 * M * K);
+    SimpleDeviceMem b0_device_buf(sizeof(B0DataType) * G0 * G1 * N * K);
+    SimpleDeviceMem d0_device_buf(sizeof(D0DataType) * G0 * G1 * M * N);
+    SimpleDeviceMem b1_device_buf(sizeof(B1DataType) * G0 * G1 * O * N);
+    SimpleDeviceMem c_device_buf(sizeof(CDataType) * G0 * G1 * M * O);
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                                          1,
+                                                                          1,
+                                                                          1,
+                                                                          1,
+                                                                          ADataType,
+                                                                          B0DataType,
+                                                                          B1DataType,
+                                                                          CDataType,
+                                                                          ck::Tuple<D0DataType>,
+                                                                          ck::Tuple<>,
+                                                                          AElementOp,
+                                                                          B0ElementOp,
+                                                                          Acc0ElementOp,
+                                                                          B1ElementOp,
+                                                                          CElementOp,
+                                                                          MaskingSpec>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b0_device_buf.GetDeviceBuffer(),
+            b1_device_buf.GetDeviceBuffer(),
+            c_device_buf.GetDeviceBuffer(),
+            std::array<void*, 1>{d0_device_buf.GetDeviceBuffer()}, // p_acc0_biases
+            {},                                                    // p_acc1_biases
+            a_gs_ms_ks_lengths,
+            a_gs_ms_ks_strides,
+            b0_gs_ns_ks_lengths,
+            b0_gs_ns_ks_strides,
+            b1_gs_os_ns_lengths,
+            b1_gs_os_ns_strides,
+            c_gs_ms_os_lengths,
+            c_gs_ms_os_strides,
+            std::array<std::vector<ck::index_t>, 1>{
+                d0_gs_ms_ns_lengths}, // acc0_biases_gs_ms_ns_lengths
+            std::array<std::vector<ck::index_t>, 1>{
+                d0_gs_ms_ns_strides}, // acc0_biases_gs_ms_ns_strides
+            {},                       // acc1_biases_gs_ms_os_lengths
+            {},                       // acc1_biases_gs_ms_os_strides
+            AElementOp{},
+            B0ElementOp{},
+            Acc0ElementOp{1 / sqrtf(K)},
+            B1ElementOp{},
+            CElementOp{});
+
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop      = (size_t(M) * N * K * 2 + size_t(M) * N * O * 2) * G0 * G1;
+            std::size_t num_btype = (sizeof(ADataType) * M * K + sizeof(B0DataType) * K * N +
+                                     sizeof(B1DataType) * N * O + sizeof(CDataType) * M * O +
+                                     sizeof(D0DataType) * M * N) *
+                                    G0 * G1;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                      << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best instance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b0_device_buf.GetDeviceBuffer(),
+            b1_device_buf.GetDeviceBuffer(),
+            c_device_buf.GetDeviceBuffer(),
+            std::array<void*, 1>{d0_device_buf.GetDeviceBuffer()}, // p_acc0_biases
+            {},                                                    // p_acc1_biases
+            a_gs_ms_ks_lengths,
+            a_gs_ms_ks_strides,
+            b0_gs_ns_ks_lengths,
+            b0_gs_ns_ks_strides,
+            b1_gs_os_ns_lengths,
+            b1_gs_os_ns_strides,
+            c_gs_ms_os_lengths,
+            c_gs_ms_os_strides,
+            std::array<std::vector<ck::index_t>, 1>{
+                d0_gs_ms_ns_lengths}, // acc0_biases_gs_ms_ns_lengths
+            std::array<std::vector<ck::index_t>, 1>{
+                d0_gs_ms_ns_strides}, // acc0_biases_gs_ms_ns_strides
+            {},                       // acc1_biases_gs_ms_os_lengths
+            {},                       // acc1_biases_gs_ms_os_strides
+            AElementOp{},
+            B0ElementOp{},
+            Acc0ElementOp{1 / sqrtf(K)},
+            B1ElementOp{},
+            CElementOp{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/09_quantization/CMakeLists.txt b/client_example/09_quantization/CMakeLists.txt
index 7dc9b860c0c427a21e6127fcce5556dbb06089e9..2b7d6fc806ad48f650c3e1f72ef3190cb9f342f4 100644
--- a/client_example/09_quantization/CMakeLists.txt
+++ b/client_example/09_quantization/CMakeLists.txt
@@ -1,6 +1,12 @@
+add_executable(client_conv2d_fwd_bias_tanh_perchannel_quantization conv2d_fwd_bias_tanh_perchannel_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_bias_tanh_perchannel_quantization PRIVATE composable_kernel::device_operations)
+
 add_executable(client_conv2d_fwd_bias_relu_perchannel_quantization conv2d_fwd_bias_relu_perchannel_quantization.cpp)
 target_link_libraries(client_conv2d_fwd_bias_relu_perchannel_quantization PRIVATE composable_kernel::device_operations)
 
+add_executable(client_conv2d_fwd_bias_tanh_perlayer_quantization conv2d_fwd_bias_tanh_perlayer_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_bias_tanh_perlayer_quantization PRIVATE composable_kernel::device_operations)
+
 add_executable(client_conv2d_fwd_bias_relu_perlayer_quantization conv2d_fwd_bias_relu_perlayer_quantization.cpp)
 target_link_libraries(client_conv2d_fwd_bias_relu_perlayer_quantization PRIVATE composable_kernel::device_operations)
 
@@ -9,3 +15,6 @@ target_link_libraries(client_conv2d_fwd_perchannel_quantization PRIVATE composab
 
 add_executable(client_conv2d_fwd_perlayer_quantization conv2d_fwd_perlayer_quantization.cpp)
 target_link_libraries(client_conv2d_fwd_perlayer_quantization PRIVATE composable_kernel::device_operations)
+
+add_executable(client_gemm_quantization gemm_quantization.cpp)
+target_link_libraries(client_gemm_quantization PRIVATE composable_kernel::device_operations)
diff --git a/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
index bcb0cefa712cf144edf1916adf0c0f97515d56f0..cd504e942e943b8f174442554eca379cd908ba6e 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
@@ -17,27 +17,26 @@ using BiasDataType         = int32_t;
 using RequantScaleDataType = float;
 using OutDataType          = int8_t;
 
-using InLayout           = ck::tensor_layout::convolution::GNHWC;
+using InLayout           = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
 using BiasLayout         = ck::tensor_layout::convolution::G_K;
 using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 using PassThrough        = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp       = ck::tensor_operation::element_wise::Relu;
 using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
-static constexpr ck::index_t N             = 4;
-static constexpr ck::index_t K             = 64;
-static constexpr ck::index_t C             = 32;
-static constexpr ck::index_t Y             = 3;
-static constexpr ck::index_t X             = 3;
-static constexpr ck::index_t Hi            = 71;
-static constexpr ck::index_t Wi            = 71;
-static constexpr ck::index_t Ho            = 36;
-static constexpr ck::index_t Wo            = 36;
-
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;  // batch size
+static constexpr ck::index_t K             = 32; // output channel
+static constexpr ck::index_t C             = 64; // input channel (per group)
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Hi            = 71; // input H
+static constexpr ck::index_t Wi            = 71; // input W
+static constexpr ck::index_t Ho            = 36; // output H
+static constexpr ck::index_t Wo            = 36; // output W
 struct SimpleDeviceMem
 {
     SimpleDeviceMem() = delete;
@@ -56,26 +55,30 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> bias_strides{K, 0, 1, 0, 0};
     std::array<ck::index_t, 5> requant_scale_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> requant_scale_strides{K, 0, 1, 0, 0};
-    std::array<ck::index_t, 5> out_lengths{G, N, C, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * C, Ho * Wo * C, 1, Wo * C, C};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem bias(sizeof(BiasDataType) * K * Y * X * C);
-    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
         NumDimSpatial,
@@ -136,10 +139,11 @@ int main(int argc, char* argv[])
         {
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
-            std::size_t flop      = G * 2 * N * K * C * Ho * Wo * Y * X;
-            std::size_t num_bytes = G * sizeof(InDataType) * N * Hi * Wi * C +
-                                    G * sizeof(WeiDataType) * K * Y * X * C +
-                                    G * sizeof(OutDataType) * N * Ho * Wo * K;
+            std::size_t flop = G * 2 * N * K * C * Ho * Wo * Y * X;
+            std::size_t num_bytes =
+                G * sizeof(InDataType) * N * Hi * Wi * C + G * sizeof(WeiDataType) * K * Y * X * C +
+                G * sizeof(BiasDataType) * K + G * sizeof(RequantScaleDataType) * K +
+                G * sizeof(OutDataType) * N * Ho * Wo * K;
 
             float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
@@ -162,11 +166,12 @@ int main(int argc, char* argv[])
         }
     }
 
-    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
-              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
-
     // run the best intance
+    if(best_op_id != -1)
     {
+        std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+                  << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
         auto& op_ptr = op_ptrs[best_op_id];
         std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
                   << std::endl;
@@ -202,4 +207,4 @@ int main(int argc, char* argv[])
     }
 
     return 0;
-}
\ No newline at end of file
+}
diff --git a/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
index 26c7aa15e2be0814b20b17f9c2c91fe21e70d961..f4aa3666b1c15eccd37bdef549f8402bcd1b252b 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
@@ -16,25 +16,26 @@ using WeiDataType  = int8_t;
 using BiasDataType = int32_t;
 using OutDataType  = int8_t;
 
-using InLayout     = ck::tensor_layout::convolution::GNHWC;
+using InLayout     = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout    = ck::tensor_layout::convolution::GKYXC;
 using BiasLayout   = ck::tensor_layout::convolution::G_K;
-using OutLayout    = ck::tensor_layout::convolution::GNHWK;
+using OutLayout    = ck::tensor_layout::convolution::NHWGK;
 using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp = ck::tensor_operation::element_wise::Relu;
 using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
-static constexpr ck::index_t N             = 4;
-static constexpr ck::index_t K             = 64;
-static constexpr ck::index_t C             = 32;
-static constexpr ck::index_t Y             = 3;
-static constexpr ck::index_t X             = 3;
-static constexpr ck::index_t Hi            = 71;
-static constexpr ck::index_t Wi            = 71;
-static constexpr ck::index_t Ho            = 36;
-static constexpr ck::index_t Wo            = 36;
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;    // batch size
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
+static constexpr ck::index_t Y             = 3;    // filter H
+static constexpr ck::index_t X             = 3;    // filter W
+static constexpr ck::index_t Hi            = 71;   // input H
+static constexpr ck::index_t Wi            = 71;   // input W
+static constexpr ck::index_t Ho            = 36;   // output H
+static constexpr ck::index_t Wo            = 36;   // output W
+static constexpr float requant_scale       = 0.5f; // requantize qAcc to qz
 
 struct SimpleDeviceMem
 {
@@ -54,23 +55,27 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> bias_strides{K, 0, 1, 0, 0};
-    std::array<ck::index_t, 5> out_lengths{G, N, C, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * C, Ho * Wo * C, 1, Wo * C, C};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem bias(sizeof(BiasDataType) * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp =
         ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
@@ -102,26 +107,27 @@ int main(int argc, char* argv[])
 
     for(int i = 0; i < op_ptrs.size(); ++i)
     {
-        auto& op_ptr      = op_ptrs[i];
-        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
-                                                        wei.GetDeviceBuffer(),
-                                                        {bias.GetDeviceBuffer()},
-                                                        out.GetDeviceBuffer(),
-                                                        in_lengths,
-                                                        in_strides,
-                                                        weight_lengths,
-                                                        weight_strides,
-                                                        {bias_lengths},
-                                                        {bias_strides},
-                                                        out_lengths,
-                                                        out_strides,
-                                                        conv_strides,
-                                                        conv_dilations,
-                                                        in_left_pad,
-                                                        in_right_pad,
-                                                        PassThrough{},
-                                                        PassThrough{},
-                                                        OutElementOp{0.5f, ActivationOp{}});
+        auto& op_ptr = op_ptrs[i];
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {bias.GetDeviceBuffer()},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        weight_lengths,
+                                        weight_strides,
+                                        {bias_lengths},
+                                        {bias_strides},
+                                        out_lengths,
+                                        out_strides,
+                                        conv_strides,
+                                        conv_dilations,
+                                        in_left_pad,
+                                        in_right_pad,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        OutElementOp{requant_scale, ActivationOp{}});
 
         auto invoker_ptr    = op_ptr->MakeInvokerPointer();
         std::string op_name = op_ptr->GetTypeString();
@@ -130,10 +136,10 @@ int main(int argc, char* argv[])
         {
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
-            std::size_t flop      = G * 2 * N * K * C * Ho * Wo * Y * X;
-            std::size_t num_bytes = G * sizeof(InDataType) * N * Hi * Wi * C +
-                                    G * sizeof(WeiDataType) * K * Y * X * C +
-                                    G * sizeof(OutDataType) * N * Ho * Wo * K;
+            std::size_t flop = G * 2 * N * K * C * Ho * Wo * Y * X;
+            std::size_t num_bytes =
+                G * sizeof(InDataType) * N * Hi * Wi * C + G * sizeof(WeiDataType) * K * Y * X * C +
+                G * sizeof(BiasDataType) * K + G * sizeof(OutDataType) * N * Ho * Wo * K;
 
             float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
@@ -156,33 +162,35 @@ int main(int argc, char* argv[])
         }
     }
 
-    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
-              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
-
     // run the best intance
+    if(best_op_id != -1)
     {
+        std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+                  << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
         auto& op_ptr = op_ptrs[best_op_id];
         std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
                   << std::endl;
-        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
-                                                        wei.GetDeviceBuffer(),
-                                                        {bias.GetDeviceBuffer()},
-                                                        out.GetDeviceBuffer(),
-                                                        in_lengths,
-                                                        in_strides,
-                                                        weight_lengths,
-                                                        weight_strides,
-                                                        {bias_lengths},
-                                                        {bias_strides},
-                                                        out_lengths,
-                                                        out_strides,
-                                                        conv_strides,
-                                                        conv_dilations,
-                                                        in_left_pad,
-                                                        in_right_pad,
-                                                        PassThrough{},
-                                                        PassThrough{},
-                                                        OutElementOp{0.5f, ActivationOp{}});
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {bias.GetDeviceBuffer()},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        weight_lengths,
+                                        weight_strides,
+                                        {bias_lengths},
+                                        {bias_strides},
+                                        out_lengths,
+                                        out_strides,
+                                        conv_strides,
+                                        conv_dilations,
+                                        in_left_pad,
+                                        in_right_pad,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        OutElementOp{requant_scale, ActivationOp{}});
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
diff --git a/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ebdbbf52c0ca257dd8f672a48b32f80fa5bb8816
--- /dev/null
+++ b/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
@@ -0,0 +1,213 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using BiasDataType         = int32_t;
+using RequantScaleDataType = float;
+using OutDataType          = int8_t;
+
+using InLayout           = ck::tensor_layout::convolution::NHWGC;
+using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
+using BiasLayout         = ck::tensor_layout::convolution::G_K;
+using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
+using OutLayout          = ck::tensor_layout::convolution::NHWGK;
+using PassThrough        = ck::tensor_operation::element_wise::PassThrough;
+using ActivationOp       = ck::tensor_operation::element_wise::TanH;
+using OutElementOp =
+    ck::tensor_operation::element_wise::Add_Mul2_Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;    // batch size
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
+static constexpr ck::index_t Y             = 3;    // filter H
+static constexpr ck::index_t X             = 3;    // filter W
+static constexpr ck::index_t Hi            = 71;   // input H
+static constexpr ck::index_t Wi            = 71;   // input W
+static constexpr ck::index_t Ho            = 36;   // output H
+static constexpr ck::index_t Wo            = 36;   // output W
+static constexpr float sz_inv              = 0.5f; // inverse of scale_z
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
+    std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
+    std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
+    std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
+    std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> bias_strides{K, 0, 1, 0, 0};
+    std::array<ck::index_t, 5> requant_scale_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> requant_scale_strides{K, 0, 1, 0, 0};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
+    std::array<ck::index_t, 2> in_left_pad{1, 1};
+    std::array<ck::index_t, 2> in_right_pad{1, 1};
+    std::array<ck::index_t, 2> conv_strides{2, 2};
+    std::array<ck::index_t, 2> conv_dilations{1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+        NumDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout, RequantScaleLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType, RequantScaleDataType>,
+        OutDataType,
+        PassThrough,
+        PassThrough,
+        OutElementOp>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {bias.GetDeviceBuffer(), requant_scale.GetDeviceBuffer()},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        weight_lengths,
+                                        weight_strides,
+                                        {bias_lengths, requant_scale_lengths},
+                                        {bias_strides, requant_scale_strides},
+                                        out_lengths,
+                                        out_strides,
+                                        conv_strides,
+                                        conv_dilations,
+                                        in_left_pad,
+                                        in_right_pad,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        OutElementOp{sz_inv, ActivationOp{}});
+
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = G * 2 * N * K * C * Ho * Wo * Y * X;
+            std::size_t num_bytes =
+                G * sizeof(InDataType) * N * Hi * Wi * C + G * sizeof(WeiDataType) * K * Y * X * C +
+                G * sizeof(BiasDataType) * K + G * sizeof(RequantScaleDataType) * K +
+                G * sizeof(OutDataType) * N * Ho * Wo * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(best_op_id != -1)
+    {
+        std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+                  << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {bias.GetDeviceBuffer(), requant_scale.GetDeviceBuffer()},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        weight_lengths,
+                                        weight_strides,
+                                        {bias_lengths, requant_scale_lengths},
+                                        {bias_strides, requant_scale_strides},
+                                        out_lengths,
+                                        out_strides,
+                                        conv_strides,
+                                        conv_dilations,
+                                        in_left_pad,
+                                        in_right_pad,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        OutElementOp{sz_inv, ActivationOp{}});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..9d60baee06fddb27451f1f51642bad738739c4bb
--- /dev/null
+++ b/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
@@ -0,0 +1,205 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType   = int8_t;
+using WeiDataType  = int8_t;
+using BiasDataType = int32_t;
+using OutDataType  = int8_t;
+
+using InLayout     = ck::tensor_layout::convolution::NHWGC;
+using WeiLayout    = ck::tensor_layout::convolution::GKYXC;
+using BiasLayout   = ck::tensor_layout::convolution::G_K;
+using OutLayout    = ck::tensor_layout::convolution::NHWGK;
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::TanH;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Mul_Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;    // batch size
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
+static constexpr ck::index_t Y             = 3;    // filter H
+static constexpr ck::index_t X             = 3;    // filter W
+static constexpr ck::index_t Hi            = 71;   // input H
+static constexpr ck::index_t Wi            = 71;   // input W
+static constexpr ck::index_t Ho            = 36;   // output H
+static constexpr ck::index_t Wo            = 36;   // output W
+static constexpr float sacc                = 0.5f; //  scale of acc
+static constexpr float sz_inv              = 0.5f; // inverse of scale_z
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
+    std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
+    std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
+    std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
+    std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> bias_strides{K, 0, 1, 0, 0};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
+    std::array<ck::index_t, 2> in_left_pad{1, 1};
+    std::array<ck::index_t, 2> in_right_pad{1, 1};
+    std::array<ck::index_t, 2> conv_strides{2, 2};
+    std::array<ck::index_t, 2> conv_dilations{1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
+                                                                    InLayout,
+                                                                    WeiLayout,
+                                                                    ck::Tuple<BiasLayout>,
+                                                                    OutLayout,
+                                                                    InDataType,
+                                                                    WeiDataType,
+                                                                    ck::Tuple<BiasDataType>,
+                                                                    OutDataType,
+                                                                    PassThrough,
+                                                                    PassThrough,
+                                                                    OutElementOp>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {bias.GetDeviceBuffer()},
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        weight_lengths,
+                                                        weight_strides,
+                                                        {bias_lengths},
+                                                        {bias_strides},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        conv_strides,
+                                                        conv_dilations,
+                                                        in_left_pad,
+                                                        in_right_pad,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        OutElementOp{sacc, sz_inv, ActivationOp{}});
+
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = G * 2 * N * K * C * Ho * Wo * Y * X;
+            std::size_t num_bytes =
+                G * sizeof(InDataType) * N * Hi * Wi * C + G * sizeof(WeiDataType) * K * Y * X * C +
+                G * sizeof(BiasDataType) * K + G * sizeof(OutDataType) * N * Ho * Wo * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(best_op_id != -1)
+    {
+        std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+                  << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {bias.GetDeviceBuffer()},
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        weight_lengths,
+                                                        weight_strides,
+                                                        {bias_lengths},
+                                                        {bias_strides},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        conv_strides,
+                                                        conv_dilations,
+                                                        in_left_pad,
+                                                        in_right_pad,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        OutElementOp{sacc, sz_inv, ActivationOp{}});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
\ No newline at end of file
diff --git a/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
index 475b2f03b4f558ad4bc319393472b926ebbfee2d..dd81d9ee6b6dc0ee453cac61d7e795c79ba3c9fa 100644
--- a/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
@@ -16,25 +16,25 @@ using WeiDataType          = int8_t;
 using RequantScaleDataType = float;
 using OutDataType          = int8_t;
 
-using InLayout           = ck::tensor_layout::convolution::GNHWC;
+using InLayout           = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
 using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 using PassThrough        = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp       = PassThrough;
 using OutElementOp       = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
-static constexpr ck::index_t N             = 4;
-static constexpr ck::index_t K             = 64;
-static constexpr ck::index_t C             = 32;
-static constexpr ck::index_t Y             = 3;
-static constexpr ck::index_t X             = 3;
-static constexpr ck::index_t Hi            = 71;
-static constexpr ck::index_t Wi            = 71;
-static constexpr ck::index_t Ho            = 36;
-static constexpr ck::index_t Wo            = 36;
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;  // batch size
+static constexpr ck::index_t K             = 32; // output channel
+static constexpr ck::index_t C             = 64; // input channel (per group)
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Hi            = 71; // input H
+static constexpr ck::index_t Wi            = 71; // input W
+static constexpr ck::index_t Ho            = 36; // output H
+static constexpr ck::index_t Wo            = 36; // output W
 
 struct SimpleDeviceMem
 {
@@ -54,23 +54,27 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> requant_scale_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> requant_scale_strides{K, 0, 1, 0, 0};
-    std::array<ck::index_t, 5> out_lengths{G, N, C, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * C, Ho * Wo * C, 1, Wo * C, C};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp =
         ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
@@ -130,10 +134,10 @@ int main(int argc, char* argv[])
         {
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
-            std::size_t flop      = G * 2 * N * K * C * Ho * Wo * Y * X;
-            std::size_t num_bytes = G * sizeof(InDataType) * N * Hi * Wi * C +
-                                    G * sizeof(WeiDataType) * K * Y * X * C +
-                                    G * sizeof(OutDataType) * N * Ho * Wo * K;
+            std::size_t flop = G * 2 * N * K * C * Ho * Wo * Y * X;
+            std::size_t num_bytes =
+                G * sizeof(InDataType) * N * Hi * Wi * C + G * sizeof(WeiDataType) * K * Y * X * C +
+                G * sizeof(RequantScaleDataType) * K + G * sizeof(OutDataType) * N * Ho * Wo * K;
 
             float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
@@ -156,11 +160,12 @@ int main(int argc, char* argv[])
         }
     }
 
-    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
-              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
-
     // run the best intance
+    if(best_op_id != -1)
     {
+        std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+                  << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
         auto& op_ptr = op_ptrs[best_op_id];
         std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
                   << std::endl;
@@ -195,4 +200,4 @@ int main(int argc, char* argv[])
     }
 
     return 0;
-}
\ No newline at end of file
+}
diff --git a/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
index da7b7e6abffd1f5033e4a02700a1a939a950dc0e..9c088a21d38e7f62c8eee35bf4c557a0bb2209bd 100644
--- a/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
@@ -15,24 +15,25 @@ using InDataType  = int8_t;
 using WeiDataType = int8_t;
 using OutDataType = int8_t;
 
-using InLayout     = ck::tensor_layout::convolution::GNHWC;
+using InLayout     = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout    = ck::tensor_layout::convolution::GKYXC;
-using OutLayout    = ck::tensor_layout::convolution::GNHWK;
+using OutLayout    = ck::tensor_layout::convolution::NHWGK;
 using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp = PassThrough;
 using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
-static constexpr ck::index_t N             = 4;
-static constexpr ck::index_t K             = 64;
-static constexpr ck::index_t C             = 32;
-static constexpr ck::index_t Y             = 3;
-static constexpr ck::index_t X             = 3;
-static constexpr ck::index_t Hi            = 71;
-static constexpr ck::index_t Wi            = 71;
-static constexpr ck::index_t Ho            = 36;
-static constexpr ck::index_t Wo            = 36;
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;    // batch size
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
+static constexpr ck::index_t Y             = 3;    // filter H
+static constexpr ck::index_t X             = 3;    // filter W
+static constexpr ck::index_t Hi            = 71;   // input H
+static constexpr ck::index_t Wi            = 71;   // input W
+static constexpr ck::index_t Ho            = 36;   // output H
+static constexpr ck::index_t Wo            = 36;   // output W
+static constexpr float requant_scale       = 0.5f; // requantize qAcc to qY
 
 struct SimpleDeviceMem
 {
@@ -52,20 +53,24 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
-    std::array<ck::index_t, 5> out_lengths{G, N, C, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * C, Ho * Wo * C, 1, Wo * C, C};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
                                                                                  InLayout,
@@ -96,26 +101,27 @@ int main(int argc, char* argv[])
 
     for(int i = 0; i < op_ptrs.size(); ++i)
     {
-        auto& op_ptr      = op_ptrs[i];
-        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
-                                                        wei.GetDeviceBuffer(),
-                                                        {},
-                                                        out.GetDeviceBuffer(),
-                                                        in_lengths,
-                                                        in_strides,
-                                                        weight_lengths,
-                                                        weight_strides,
-                                                        {},
-                                                        {},
-                                                        out_lengths,
-                                                        out_strides,
-                                                        conv_strides,
-                                                        conv_dilations,
-                                                        in_left_pad,
-                                                        in_right_pad,
-                                                        PassThrough{},
-                                                        PassThrough{},
-                                                        OutElementOp{0.5f, ActivationOp{}});
+        auto& op_ptr = op_ptrs[i];
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        weight_lengths,
+                                        weight_strides,
+                                        {},
+                                        {},
+                                        out_lengths,
+                                        out_strides,
+                                        conv_strides,
+                                        conv_dilations,
+                                        in_left_pad,
+                                        in_right_pad,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        OutElementOp{requant_scale, ActivationOp{}});
 
         auto invoker_ptr    = op_ptr->MakeInvokerPointer();
         std::string op_name = op_ptr->GetTypeString();
@@ -150,33 +156,34 @@ int main(int argc, char* argv[])
         }
     }
 
-    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
-              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
-
-    // run the best intance
+    if(best_op_id != -1)
     {
+        std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+                  << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
         auto& op_ptr = op_ptrs[best_op_id];
         std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
                   << std::endl;
-        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
-                                                        wei.GetDeviceBuffer(),
-                                                        {},
-                                                        out.GetDeviceBuffer(),
-                                                        in_lengths,
-                                                        in_strides,
-                                                        weight_lengths,
-                                                        weight_strides,
-                                                        {},
-                                                        {},
-                                                        out_lengths,
-                                                        out_strides,
-                                                        conv_strides,
-                                                        conv_dilations,
-                                                        in_left_pad,
-                                                        in_right_pad,
-                                                        PassThrough{},
-                                                        PassThrough{},
-                                                        OutElementOp{0.5f, ActivationOp{}});
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        weight_lengths,
+                                        weight_strides,
+                                        {},
+                                        {},
+                                        out_lengths,
+                                        out_strides,
+                                        conv_strides,
+                                        conv_dilations,
+                                        in_left_pad,
+                                        in_right_pad,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        OutElementOp{requant_scale, ActivationOp{}});
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
diff --git a/client_example/09_quantization/gemm_quantization.cpp b/client_example/09_quantization/gemm_quantization.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b14e68fa082f8f9d05ab5f00471a2fa82d1d113b
--- /dev/null
+++ b/client_example/09_quantization/gemm_quantization.cpp
@@ -0,0 +1,193 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp"
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using ActivationOp = PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
+
+using ADataType = int8_t;
+using BDataType = int8_t;
+using EDataType = int8_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using ELayout = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t M = 1024;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
+
+    ck::index_t StrideA = 1024;
+    ck::index_t StrideB = 1024;
+    ck::index_t StrideE = 1024;
+
+    float requant_scale = 0.03;
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<ALayout,
+                                                                       BLayout,
+                                                                       ck::Tuple<>,
+                                                                       ELayout,
+                                                                       ADataType,
+                                                                       BDataType,
+                                                                       ck::Tuple<>,
+                                                                       EDataType,
+                                                                       AElementOp,
+                                                                       BElementOp,
+                                                                       CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{requant_scale, ActivationOp{}};
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                                        b_device_buf.GetDeviceBuffer(),
+                                                        {},
+                                                        e_device_buf.GetDeviceBuffer(),
+                                                        M,
+                                                        N,
+                                                        K,
+                                                        StrideA,
+                                                        StrideB,
+                                                        {},
+                                                        StrideE,
+                                                        a_element_op,
+                                                        b_element_op,
+                                                        cde_element_op);
+
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = std::size_t(2) * M * N * K;
+            std::size_t num_bytes =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id != -1)
+    {
+        std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+                  << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                                        b_device_buf.GetDeviceBuffer(),
+                                                        {},
+                                                        e_device_buf.GetDeviceBuffer(),
+                                                        M,
+                                                        N,
+                                                        K,
+                                                        StrideA,
+                                                        StrideB,
+                                                        {},
+                                                        StrideE,
+                                                        a_element_op,
+                                                        b_element_op,
+                                                        cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
\ No newline at end of file
diff --git a/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp b/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp
index 55c789804230ccccf66d68be9244c5c4111451e6..1b2e8abc201c2aed2cd2eebccb68405a25033a43 100644
--- a/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp
+++ b/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iomanip>
diff --git a/client_example/11_grouped_conv_bwd_weight/CMakeLists.txt b/client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
index 3e3f6677666545a616a5664080c7fd20ac8ae4e0..82162b606a6ee055cb985b960366a93fa4dd21e3 100644
--- a/client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/client_example/11_grouped_conv_bwd_weight/CMakeLists.txt
@@ -1,2 +1,9 @@
-add_executable(client_grouped_conv2d_bwd_weight grouped_conv2d_bwd_weight.cpp)
-target_link_libraries(client_grouped_conv2d_bwd_weight PRIVATE composable_kernel::device_operations)
+add_executable(client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_fp16.cpp)
+add_executable(client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp)
+add_executable(client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp)
+add_executable(client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp)
+
+target_link_libraries(client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_operations)
diff --git a/client_example/11_grouped_conv_bwd_weight/common.hpp b/client_example/11_grouped_conv_bwd_weight/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..f63e5f2157646ad2003fbe3bbdcc10f4a96e1d7e
--- /dev/null
+++ b/client_example/11_grouped_conv_bwd_weight/common.hpp
@@ -0,0 +1,252 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+template <ck::index_t NumDimSpatial>
+std::size_t GetFlops(ck::index_t G,
+                     ck::index_t N,
+                     ck::index_t K,
+                     ck::index_t C,
+                     const std::array<ck::index_t, NumDimSpatial>& output_spatial_lengths,
+                     const std::array<ck::index_t, NumDimSpatial>& filter_spatial_lengths)
+{
+    // 2 * G * N * K * C * <output spatial lengths product> * <filter spatial lengths product>
+    return static_cast<std::size_t>(2) * G * N * K * C *
+           std::accumulate(std::begin(output_spatial_lengths),
+                           std::end(output_spatial_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>()) *
+           std::accumulate(std::begin(filter_spatial_lengths),
+                           std::end(filter_spatial_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>());
+}
+
+template <typename InDataType, ck::index_t NumDimSpatial>
+std::size_t GetInputByte(ck::index_t G,
+                         ck::index_t N,
+                         ck::index_t C,
+                         const std::array<ck::index_t, NumDimSpatial>& input_spatial_lengths)
+{
+    // sizeof(InDataType) * (G * N * C * <input spatial lengths product>) +
+    return sizeof(InDataType) * (G * N * C *
+                                 std::accumulate(std::begin(input_spatial_lengths),
+                                                 std::end(input_spatial_lengths),
+                                                 static_cast<std::size_t>(1),
+                                                 std::multiplies<>()));
+}
+
+template <typename WeiDataType, ck::index_t NumDimSpatial>
+std::size_t GetWeightByte(ck::index_t G,
+                          ck::index_t K,
+                          ck::index_t C,
+                          const std::array<ck::index_t, NumDimSpatial>& filter_spatial_lengths)
+{
+    // sizeof(WeiDataType) * (G * K * C * <filter spatial lengths product>) +
+    return sizeof(WeiDataType) * (G * K * C *
+                                  std::accumulate(std::begin(filter_spatial_lengths),
+                                                  std::end(filter_spatial_lengths),
+                                                  static_cast<std::size_t>(1),
+                                                  std::multiplies<>()));
+}
+
+template <typename OutDataType, ck::index_t NumDimSpatial>
+std::size_t GetOutputByte(ck::index_t G,
+                          ck::index_t N,
+                          ck::index_t K,
+                          const std::array<ck::index_t, NumDimSpatial>& output_spatial_lengths)
+{
+    // sizeof(OutDataType) * (G * N * K * <output spatial lengths product>);
+    return sizeof(OutDataType) * (G * N * K *
+                                  std::accumulate(std::begin(output_spatial_lengths),
+                                                  std::end(output_spatial_lengths),
+                                                  static_cast<std::size_t>(1),
+                                                  std::multiplies<std::size_t>()));
+}
+
+template <ck::index_t NumDimSpatial,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout>
+bool run_grouped_conv_bwd_weight(
+    const ck::index_t G,
+    const ck::index_t N,
+    const ck::index_t K,
+    const ck::index_t C,
+    const std::array<ck::index_t, NumDimSpatial>& input_spatial_lengths,
+    const std::array<ck::index_t, NumDimSpatial>& filter_spatial_lengths,
+    const std::array<ck::index_t, NumDimSpatial>& output_spatial_lengths,
+    const std::array<ck::index_t, NumDimSpatial + 3>& input_strides,
+    const std::array<ck::index_t, NumDimSpatial + 3>& output_strides,
+    const std::array<ck::index_t, NumDimSpatial>& conv_filter_strides,
+    const std::array<ck::index_t, NumDimSpatial>& conv_filter_dilations,
+    const std::array<ck::index_t, NumDimSpatial>& input_left_pads,
+    const std::array<ck::index_t, NumDimSpatial>& input_right_pads)
+{
+
+    ck::index_t split_k = 2;
+    SimpleDeviceMem in(GetInputByte<InDataType, NumDimSpatial>(G, N, C, input_spatial_lengths));
+    SimpleDeviceMem wei(GetWeightByte<WeiDataType, NumDimSpatial>(G, K, C, filter_spatial_lengths));
+    SimpleDeviceMem out(GetOutputByte<OutDataType, NumDimSpatial>(G, N, K, output_spatial_lengths));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdWeight<NumDimSpatial,
+                                                                              InLayout,
+                                                                              WeiLayout,
+                                                                              OutLayout,
+                                                                              InDataType,
+                                                                              WeiDataType,
+                                                                              OutDataType,
+                                                                              PassThrough,
+                                                                              PassThrough,
+                                                                              PassThrough>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        G,
+                                                        N,
+                                                        K,
+                                                        C,
+                                                        input_spatial_lengths,
+                                                        filter_spatial_lengths,
+                                                        output_spatial_lengths,
+                                                        input_strides,
+                                                        output_strides,
+                                                        conv_filter_strides,
+                                                        conv_filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        split_k);
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop =
+                GetFlops<NumDimSpatial>(G, N, K, C, output_spatial_lengths, filter_spatial_lengths);
+            std::size_t num_bytes =
+                GetInputByte<InDataType, NumDimSpatial>(G, N, C, input_spatial_lengths) +
+                GetWeightByte<WeiDataType, NumDimSpatial>(G, K, C, filter_spatial_lengths) +
+                GetOutputByte<OutDataType, NumDimSpatial>(G, N, K, output_spatial_lengths);
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return false;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        G,
+                                                        N,
+                                                        K,
+                                                        C,
+                                                        input_spatial_lengths,
+                                                        filter_spatial_lengths,
+                                                        output_spatial_lengths,
+                                                        input_strides,
+                                                        output_strides,
+                                                        conv_filter_strides,
+                                                        conv_filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        split_k);
+        auto invoker_ptr  = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return true;
+}
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1c6f485da2100982c9e8f2d66b63a0c3794a49c5
--- /dev/null
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
@@ -0,0 +1,58 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout  = ck::tensor_layout::convolution::GNWC;
+using WeiLayout = ck::tensor_layout::convolution::GKXC;
+using OutLayout = ck::tensor_layout::convolution::GNWK;
+
+static constexpr ck::index_t NumDimSpatial = 1;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 256;
+static constexpr ck::index_t K             = 192;
+static constexpr ck::index_t C             = 192;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Wi            = 28;
+static constexpr ck::index_t Wo            = 28;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{N * Wi * C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{N * Wo * K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1};
+
+int main()
+{
+    return run_grouped_conv_bwd_weight<NumDimSpatial,
+                                       InDataType,
+                                       WeiDataType,
+                                       OutDataType,
+                                       InLayout,
+                                       WeiLayout,
+                                       OutLayout>(G,
+                                                  N,
+                                                  K,
+                                                  C,
+                                                  input_spatial_lengths,
+                                                  filter_spatial_lengths,
+                                                  output_spatial_lengths,
+                                                  input_strides,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..25e82f389688415fc2d86b896bc41b149e26b88a
--- /dev/null
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout  = ck::tensor_layout::convolution::GNHWC;
+using WeiLayout = ck::tensor_layout::convolution::GKYXC;
+using OutLayout = ck::tensor_layout::convolution::GNHWK;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 256;
+static constexpr ck::index_t K             = 192;
+static constexpr ck::index_t C             = 192;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 28;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
+    N * Hi * Wi * C, Hi* Wi* C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
+    N * Ho * Wo * K, Ho* Wo* K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
+
+int main()
+{
+    return run_grouped_conv_bwd_weight<NumDimSpatial,
+                                       InDataType,
+                                       WeiDataType,
+                                       OutDataType,
+                                       InLayout,
+                                       WeiLayout,
+                                       OutLayout>(G,
+                                                  N,
+                                                  K,
+                                                  C,
+                                                  input_spatial_lengths,
+                                                  filter_spatial_lengths,
+                                                  output_spatial_lengths,
+                                                  input_strides,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a5f5e628ff2d40b55a9bccd81a743b7b0f026fc9
--- /dev/null
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout  = ck::tensor_layout::convolution::GNDHWC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::GNDHWK;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 8;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 128;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 3;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 3;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Di, Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Z, Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Do, Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
+    N * Di * Hi * Wi * C, Di* Hi* Wi* C, Hi* Wi* C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
+    N * Do * Ho * Wo * K, Do* Ho* Wo* K, Ho* Wo* K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+int main()
+{
+    return run_grouped_conv_bwd_weight<NumDimSpatial,
+                                       InDataType,
+                                       WeiDataType,
+                                       OutDataType,
+                                       InLayout,
+                                       WeiLayout,
+                                       OutLayout>(G,
+                                                  N,
+                                                  K,
+                                                  C,
+                                                  input_spatial_lengths,
+                                                  filter_spatial_lengths,
+                                                  output_spatial_lengths,
+                                                  input_strides,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d95e8a205e6fb203beccb9691615c14bfb49c1fb
--- /dev/null
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
@@ -0,0 +1,67 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = float;
+using WeiDataType = float;
+using OutDataType = float;
+
+using InLayout  = ck::tensor_layout::convolution::GNDHWC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::GNDHWK;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 8;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 128;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 3;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 3;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Di, Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Z, Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Do, Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
+    N * Di * Hi * Wi * C, Di* Hi* Wi* C, Hi* Wi* C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
+    N * Do * Ho * Wo * K, Do* Ho* Wo* K, Ho* Wo* K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+int main()
+{
+    return run_grouped_conv_bwd_weight<NumDimSpatial,
+                                       InDataType,
+                                       WeiDataType,
+                                       OutDataType,
+                                       InLayout,
+                                       WeiLayout,
+                                       OutLayout>(
+               G,
+               N,
+               K,
+               C,
+               {Di, Hi, Wi},
+               {Z, Y, X},
+               {Do, Ho, Wo},
+               {N * Di * Hi * Wi * C, Di * Hi * Wi * C, Hi * Wi * C, Wi * C, C, 1},
+               {N * Do * Ho * Wo * K, Do * Ho * Wo * K, Ho * Wo * K, Wo * K, K, 1},
+               {1, 1, 1},
+               {1, 1, 1},
+               {1, 1, 1},
+               {1, 1, 1})
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp b/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp
index de68f46d398958917e49bf14178f66414590ed86..bc4a6fe0bfa9e118bbd6ba32ecc7dd68f3b8b2c3 100644
--- a/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp
+++ b/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/13_batchnorm/CMakeLists.txt b/client_example/13_batchnorm/CMakeLists.txt
index 54669678ae6f107d87eb6b979a3da7761d9b25d5..fc4f9d395c4a8b45e34db2ad0c5622484d379dba 100644
--- a/client_example/13_batchnorm/CMakeLists.txt
+++ b/client_example/13_batchnorm/CMakeLists.txt
@@ -1,4 +1,6 @@
 add_executable(client_batchnorm_fwd_nhwc batchnorm_fwd_nhwc.cpp)
 add_executable(client_batchnorm_bwd_nhwc batchnorm_bwd_nhwc.cpp)
+add_executable(client_batchnorm_infer_nhwc batchnorm_infer_nhwc.cpp)
 target_link_libraries(client_batchnorm_fwd_nhwc PRIVATE composable_kernel::device_operations)
 target_link_libraries(client_batchnorm_bwd_nhwc PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_batchnorm_infer_nhwc PRIVATE composable_kernel::device_operations)
diff --git a/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp b/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
index 8ef21986a4d9a5f1eb25e21a1073c4cc341da88d..c0140f71c15993da2e0d4b629321a355c74f004a 100644
--- a/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
+++ b/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp b/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
index 322667a46bacae8d0c681939c3890ef9ff476b0e..365373343668409f130f642d8e217cfd48a9afac 100644
--- a/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
+++ b/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/13_batchnorm/batchnorm_infer_nhwc.cpp b/client_example/13_batchnorm/batchnorm_infer_nhwc.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..5e6627ce14d113224c7b0acb4fea69cb36c1f369
--- /dev/null
+++ b/client_example/13_batchnorm/batchnorm_infer_nhwc.cpp
@@ -0,0 +1,189 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <functional>
+#include <numeric>
+#include <iomanip>
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp"
+
+using XDataType       = float;
+using YDataType       = float;
+using ScaleDataType   = float;
+using BiasDataType    = float;
+using MeanVarDataType = float;
+
+constexpr int Rank                  = 4;
+constexpr int NumBatchNormReduceDim = 3;
+
+using Normalize = ck::tensor_operation::element_wise::NormalizeInInfer;
+
+const double epsilon = std::numeric_limits<float>::epsilon();
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    std::array<ck::index_t, Rank> xyLengths{16, 8, 128, 256};
+    std::array<ck::index_t, Rank> xyStrides{8 * 128 * 256, 128 * 256, 256, 1};
+    std::array<ck::index_t, Rank - NumBatchNormReduceDim> scaleBiasMeanVarLengths{256};
+    std::array<ck::index_t, Rank - NumBatchNormReduceDim> scaleBiasMeanVarStrides{1};
+    std::array<int, NumBatchNormReduceDim> reduceDims{0, 1, 2};
+    std::array<int, Rank - NumBatchNormReduceDim> invariantDims{3};
+
+    ck::index_t numXYElement =
+        std::accumulate(xyLengths.begin(), xyLengths.end(), 1, std::multiplies<ck::index_t>());
+
+    ck::index_t numScaleBiasMeanVarElement = std::accumulate(scaleBiasMeanVarLengths.begin(),
+                                                             scaleBiasMeanVarLengths.end(),
+                                                             1,
+                                                             std::multiplies<ck::index_t>());
+
+    SimpleDeviceMem x(sizeof(XDataType) * numXYElement);
+    SimpleDeviceMem y(sizeof(YDataType) * numXYElement);
+    SimpleDeviceMem scale(sizeof(ScaleDataType) * numScaleBiasMeanVarElement);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * numScaleBiasMeanVarElement);
+    SimpleDeviceMem mean(sizeof(MeanVarDataType) * numScaleBiasMeanVarElement);
+    SimpleDeviceMem variance(sizeof(MeanVarDataType) * numScaleBiasMeanVarElement);
+
+    // values in variance need be non-negative
+    (void)hipMemset(
+        variance.GetDeviceBuffer(), 0, sizeof(MeanVarDataType) * numScaleBiasMeanVarElement);
+
+    std::array<ck::index_t, Rank> aligned_scaleBiasMeanVarStrides{0};
+
+    int i = 0;
+    for(auto dim : invariantDims)
+    {
+        assert(xyLengths[dim] == scaleBiasMeanVarLengths[i]);
+
+        aligned_scaleBiasMeanVarStrides[dim] = scaleBiasMeanVarStrides[i];
+        i++;
+    };
+
+    using DeviceOp = ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<XDataType, MeanVarDataType, MeanVarDataType, ScaleDataType, BiasDataType>,
+        ck::Tuple<YDataType>,
+        Normalize,
+        Rank>;
+
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(xyLengths,
+                                                        {xyStrides,
+                                                         aligned_scaleBiasMeanVarStrides,
+                                                         aligned_scaleBiasMeanVarStrides,
+                                                         aligned_scaleBiasMeanVarStrides,
+                                                         aligned_scaleBiasMeanVarStrides},
+                                                        {xyStrides},
+                                                        {x.GetDeviceBuffer(),
+                                                         mean.GetDeviceBuffer(),
+                                                         variance.GetDeviceBuffer(),
+                                                         scale.GetDeviceBuffer(),
+                                                         bias.GetDeviceBuffer()},
+                                                        {y.GetDeviceBuffer()},
+                                                        Normalize{epsilon});
+
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes =
+                numXYElement * (sizeof(XDataType) + sizeof(YDataType)) +
+                numScaleBiasMeanVarElement * (sizeof(ScaleDataType) + sizeof(BiasDataType) +
+                                              sizeof(MeanVarDataType) + sizeof(MeanVarDataType));
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        // run the best intance
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(xyLengths,
+                                                        {xyStrides,
+                                                         aligned_scaleBiasMeanVarStrides,
+                                                         aligned_scaleBiasMeanVarStrides,
+                                                         aligned_scaleBiasMeanVarStrides,
+                                                         aligned_scaleBiasMeanVarStrides},
+                                                        {xyStrides},
+                                                        {x.GetDeviceBuffer(),
+                                                         mean.GetDeviceBuffer(),
+                                                         variance.GetDeviceBuffer(),
+                                                         scale.GetDeviceBuffer(),
+                                                         bias.GetDeviceBuffer()},
+                                                        {y.GetDeviceBuffer()},
+                                                        Normalize{epsilon});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp b/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp
index 9cfeee1cfe106e69f83dc0184f3956d6751a2947..d45782d8e0ff37027a204c6820447286581f1138 100644
--- a/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp
+++ b/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/15_convnd_bwd_data/CMakeLists.txt b/client_example/15_convnd_bwd_data/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8a60a71674f84f6005deadf3744b7df346c0ac3c
--- /dev/null
+++ b/client_example/15_convnd_bwd_data/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_executable(client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp)
+add_executable(client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp)
+
+target_link_libraries(client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_operations)
diff --git a/client_example/15_convnd_bwd_data/common.hpp b/client_example/15_convnd_bwd_data/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..9799fb73a5a05225f053de3305de736c485acf3a
--- /dev/null
+++ b/client_example/15_convnd_bwd_data/common.hpp
@@ -0,0 +1,233 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <string>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+std::size_t GetFlops(ck::index_t N,
+                     ck::index_t K,
+                     ck::index_t C,
+                     const std::vector<ck::index_t>& output_spatial_lengths,
+                     const std::vector<ck::index_t>& weights_spatial_lengths)
+{
+    // 2 * N * K * C * <output spatial lengths product> * <filter spatial lengths product>
+
+    return static_cast<std::size_t>(2) * N * K * C *
+           std::accumulate(std::begin(output_spatial_lengths),
+                           std::end(output_spatial_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>()) *
+           std::accumulate(std::begin(weights_spatial_lengths),
+                           std::end(weights_spatial_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>());
+}
+
+template <typename InDataType>
+std::size_t
+GetInputByte(ck::index_t N, ck::index_t C, const std::vector<ck::index_t>& input_spatial_lengths)
+{
+    // sizeof(InDataType) * (N * C * <input spatial lengths product>) +
+    return sizeof(InDataType) * N * C *
+           std::accumulate(std::begin(input_spatial_lengths),
+                           std::end(input_spatial_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>());
+}
+
+template <typename WeiDataType>
+std::size_t
+GetWeightByte(ck::index_t K, ck::index_t C, const std::vector<ck::index_t>& weights_spatial_lengths)
+{
+    // sizeof(WeiDataType) * (K * C * <filter spatial lengths product>) +
+    return sizeof(WeiDataType) * K * C *
+           std::accumulate(std::begin(weights_spatial_lengths),
+                           std::end(weights_spatial_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>());
+}
+
+template <typename OutDataType>
+std::size_t
+GetOutputByte(ck::index_t N, ck::index_t K, const std::vector<ck::index_t>& output_spatial_lengths)
+{
+    // sizeof(OutDataType) * (N * K * <output spatial lengths product>);
+    return sizeof(OutDataType) * N * K *
+           std::accumulate(std::begin(output_spatial_lengths),
+                           std::end(output_spatial_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<std::size_t>());
+}
+
+template <ck::index_t NumDimSpatial,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout>
+bool run_conv_bwd_data(ck::index_t N,
+                       ck::index_t K,
+                       ck::index_t C,
+                       const std::vector<ck::index_t>& in_spatial_lengths,
+                       const std::vector<ck::index_t>& wei_spatial_lengths,
+                       const std::vector<ck::index_t>& out_spatial_lengths)
+{
+    std::size_t in_mem_size  = GetInputByte<InDataType>(N, C, in_spatial_lengths);
+    std::size_t wei_mem_size = GetWeightByte<WeiDataType>(K, C, wei_spatial_lengths);
+    std::size_t out_mem_size = GetOutputByte<OutDataType>(N, K, out_spatial_lengths);
+
+    SimpleDeviceMem in(in_mem_size);
+    SimpleDeviceMem wei(wei_mem_size);
+    SimpleDeviceMem out(out_mem_size);
+
+    std::vector<ck::index_t> filter_strides(NumDimSpatial, 1);
+    std::vector<ck::index_t> filter_dilations(NumDimSpatial, 1);
+    std::vector<ck::index_t> input_left_pads(NumDimSpatial, 1);
+    std::vector<ck::index_t> input_right_pads(NumDimSpatial, 1);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceConvBwdData<NumDimSpatial,
+                                                                     InLayout,
+                                                                     WeiLayout,
+                                                                     OutLayout,
+                                                                     InDataType,
+                                                                     WeiDataType,
+                                                                     OutDataType,
+                                                                     PassThrough,
+                                                                     PassThrough,
+                                                                     PassThrough>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    std::size_t flop      = GetFlops(N, K, C, out_spatial_lengths, wei_spatial_lengths);
+    std::size_t num_bytes = in_mem_size + wei_mem_size + out_mem_size;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        K,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        wei_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return false;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        K,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        wei_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+    return true;
+}
diff --git a/client_example/15_convnd_bwd_data/conv3d_bwd_data_fp16.cpp b/client_example/15_convnd_bwd_data/conv3d_bwd_data_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..5210567241ebd9ce84b8b0b576b295594bf8578b
--- /dev/null
+++ b/client_example/15_convnd_bwd_data/conv3d_bwd_data_fp16.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWC;
+using WeiLayout = ck::tensor_layout::convolution::KZYXC;
+using OutLayout = ck::tensor_layout::convolution::NDHWK;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 64;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 28;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 28;
+
+int main()
+{
+    return run_conv_bwd_data<NumDimSpatial,
+                             InDataType,
+                             WeiDataType,
+                             OutDataType,
+                             InLayout,
+                             WeiLayout,
+                             OutLayout>(N, K, C, {Di, Hi, Wi}, {Z, Y, X}, {Do, Ho, Wo})
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/15_convnd_bwd_data/conv3d_bwd_data_fp32.cpp b/client_example/15_convnd_bwd_data/conv3d_bwd_data_fp32.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..441bdfe7bec3bd12fe948c2efc892ad5cc4d0184
--- /dev/null
+++ b/client_example/15_convnd_bwd_data/conv3d_bwd_data_fp32.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = float;
+using WeiDataType = float;
+using OutDataType = float;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWC;
+using WeiLayout = ck::tensor_layout::convolution::KZYXC;
+using OutLayout = ck::tensor_layout::convolution::NDHWK;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 64;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 28;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 28;
+
+int main()
+{
+    return run_conv_bwd_data<NumDimSpatial,
+                             InDataType,
+                             WeiDataType,
+                             OutDataType,
+                             InLayout,
+                             WeiLayout,
+                             OutLayout>(N, K, C, {Di, Hi, Wi}, {Z, Y, X}, {Do, Ho, Wo})
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/15_gemm_add_multiply/CMakeLists.txt b/client_example/15_gemm_add_multiply/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fd2dcf9614016b93b43a52c2940d1701ee623584
--- /dev/null
+++ b/client_example/15_gemm_add_multiply/CMakeLists.txt
@@ -0,0 +1,3 @@
+
+add_executable(client_gemm_add_multiply gemm_add_multiply.cpp)
+target_link_libraries(client_gemm_add_multiply PRIVATE composable_kernel::device_operations)
\ No newline at end of file
diff --git a/client_example/15_gemm_add_multiply/gemm_add_multiply.cpp b/client_example/15_gemm_add_multiply/gemm_add_multiply.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c74d7c6bd8cf9cec54168792a8718768827d7b33
--- /dev/null
+++ b/client_example/15_gemm_add_multiply/gemm_add_multiply.cpp
@@ -0,0 +1,241 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddMultiply = ck::tensor_operation::element_wise::AddMultiply;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddMultiply;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using D0DataType = F16;
+using D1DataType = F16;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using D0Layout = Row;
+using D1Layout = Row;
+using ELayout  = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA  = 4096;
+    ck::index_t StrideB  = 4096;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideD1 = 4096;
+    ck::index_t StrideE  = 4096;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 9)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA  = std::stoi(argv[4]);
+        StrideB  = std::stoi(argv[5]);
+        StrideD0 = std::stoi(argv[6]);
+        StrideD1 = std::stoi(argv[7]);
+        StrideE  = std::stoi(argv[8]);
+    }
+    else
+    {
+        printf("arg1 to 8: M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideE\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem d0_m_n_device_buf(sizeof(D0DataType) *
+                                      f_matrix_space_size(M, N, StrideD0, D0Layout{}));
+    SimpleDeviceMem d1_m_n_device_buf(sizeof(D1DataType) *
+                                      f_matrix_space_size(M, N, StrideD1, D1Layout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceGemmMultipleD<ALayout,
+                                                          BLayout,
+                                                          ck::Tuple<D0Layout, D1Layout>,
+                                                          ELayout,
+                                                          ADataType,
+                                                          BDataType,
+                                                          ck::Tuple<D0DataType, D1DataType>,
+                                                          EDataType,
+                                                          AElementOp,
+                                                          BElementOp,
+                                                          CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b_device_buf.GetDeviceBuffer(),
+            std::array<const void*, 2>{d0_m_n_device_buf.GetDeviceBuffer(),
+                                       d1_m_n_device_buf.GetDeviceBuffer()},
+            e_device_buf.GetDeviceBuffer(),
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            std::array<ck::index_t, 2>{StrideD0, StrideD1},
+            StrideE,
+            a_element_op,
+            b_element_op,
+            cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = std::size_t(2) * M * N * K;
+
+            std::size_t num_btype =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b_device_buf.GetDeviceBuffer(),
+            std::array<const void*, 2>{d0_m_n_device_buf.GetDeviceBuffer(),
+                                       d1_m_n_device_buf.GetDeviceBuffer()},
+            e_device_buf.GetDeviceBuffer(),
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            std::array<ck::index_t, 2>{StrideD0, StrideD1},
+            StrideE,
+            a_element_op,
+            b_element_op,
+            cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/15_reduce/CMakeLists.txt b/client_example/15_reduce/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d52675ba834e666d0a480d5c778baeacb223a24a
--- /dev/null
+++ b/client_example/15_reduce/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_executable(client_reduce_nhwc_c reduce_nhwc_c.cpp)
+target_link_libraries(client_reduce_nhwc_c PRIVATE composable_kernel::device_operations)
diff --git a/client_example/15_reduce/reduce_nhwc_c.cpp b/client_example/15_reduce/reduce_nhwc_c.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b45b72f0de0199daa88e9e42be320e2669398dfe
--- /dev/null
+++ b/client_example/15_reduce/reduce_nhwc_c.cpp
@@ -0,0 +1,175 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <functional>
+#include <numeric>
+#include <iomanip>
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_reduce.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp"
+
+using InDataType  = float;
+using OutDataType = float;
+using AccDataType = float;
+using ReduceAdd   = ck::reduce::Add;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using UnaryDivide = ck::tensor_operation::element_wise::UnaryDivide;
+
+constexpr bool PropagateNan = false;
+constexpr bool OutputIndex  = false;
+
+constexpr int Rank         = 4;
+constexpr int NumReduceDim = 3;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    std::array<ck::index_t, Rank> in_lengths{16, 8, 128, 256};
+    std::array<ck::index_t, Rank> in_strides{8 * 128 * 256, 128 * 256, 256, 1};
+    std::array<ck::index_t, Rank - NumReduceDim> out_lengths{256};
+    std::array<ck::index_t, Rank - NumReduceDim> out_strides{1};
+    std::array<int, NumReduceDim> reduce_dims{0, 1, 2};
+
+    ck::index_t num_in_elements =
+        std::accumulate(in_lengths.begin(), in_lengths.end(), 1, std::multiplies<ck::index_t>());
+
+    ck::index_t num_out_elements =
+        std::accumulate(out_lengths.begin(), out_lengths.end(), 1, std::multiplies<ck::index_t>());
+
+    ck::index_t reduce_length = 1;
+
+    for(auto dim : reduce_dims)
+        reduce_length *= in_lengths[dim];
+
+    double alpha{1.0};
+    double beta{0.0};
+
+    SimpleDeviceMem in(sizeof(InDataType) * num_in_elements);
+    SimpleDeviceMem out(sizeof(OutDataType) * num_out_elements);
+
+    using DeviceOp     = ck::tensor_operation::device::DeviceReduce<InDataType,
+                                                                AccDataType,
+                                                                OutDataType,
+                                                                Rank,
+                                                                NumReduceDim,
+                                                                ReduceAdd,
+                                                                PassThrough,
+                                                                UnaryDivide,
+                                                                PropagateNan,
+                                                                OutputIndex>;
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in_lengths,
+                                                        in_strides,
+                                                        out_lengths,
+                                                        out_strides,
+                                                        reduce_dims,
+                                                        alpha,
+                                                        beta,
+                                                        in.GetDeviceBuffer(),
+                                                        nullptr,
+                                                        out.GetDeviceBuffer(),
+                                                        nullptr,
+                                                        PassThrough{},
+                                                        UnaryDivide{reduce_length});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes = num_in_elements * sizeof(InDataType) +
+                                    (beta == 0.0f ? 1 : 2) * num_out_elements * sizeof(OutDataType);
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+              << best_op_name << std::endl;
+
+    // run the best intance
+    if(found)
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in_lengths,
+                                                        in_strides,
+                                                        out_lengths,
+                                                        out_strides,
+                                                        reduce_dims,
+                                                        alpha,
+                                                        beta,
+                                                        in.GetDeviceBuffer(),
+                                                        nullptr,
+                                                        out.GetDeviceBuffer(),
+                                                        nullptr,
+                                                        PassThrough{},
+                                                        UnaryDivide{reduce_length});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/16_convnd_fwd/CMakeLists.txt b/client_example/16_convnd_fwd/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e2580a370ca480e0b9ec101e5715ac9a631a72f6
--- /dev/null
+++ b/client_example/16_convnd_fwd/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_executable(client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp)
+add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp)
+
+target_link_libraries(client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_operations)
diff --git a/client_example/16_convnd_fwd/common.hpp b/client_example/16_convnd_fwd/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..449c9466e829baa8658f17fe8e9ae86c5a297238
--- /dev/null
+++ b/client_example/16_convnd_fwd/common.hpp
@@ -0,0 +1,300 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <string>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+template <ck::index_t NumDimSpatial, ck::index_t NumNonSpatialDim = 3>
+std::size_t
+GetFlops(const std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>& output_lengths,
+         const std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>& weights_lengths)
+{
+    // 2 * G * N * K * C * <output spatial lengths product> * <filter spatial lengths product>
+    ck::index_t G = weights_lengths[0];
+    ck::index_t N = output_lengths[1];
+    ck::index_t K = weights_lengths[1];
+    ck::index_t C = weights_lengths[2];
+
+    return static_cast<std::size_t>(2) * G * N * K * C *
+           std::accumulate(std::next(std::begin(output_lengths), NumNonSpatialDim),
+                           std::end(output_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>()) *
+           std::accumulate(std::next(std::begin(weights_lengths), NumNonSpatialDim),
+                           std::end(weights_lengths),
+                           static_cast<std::size_t>(1),
+                           std::multiplies<>());
+}
+
+template <typename InDataType, ck::index_t NumDimSpatial, ck::index_t NumNonSpatialDim = 3>
+std::size_t
+GetInputByte(const std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>& input_lengths)
+{
+    // sizeof(InDataType) * (G * N * C * <input spatial lengths product>) +
+    return sizeof(InDataType) * std::accumulate(std::begin(input_lengths),
+                                                std::end(input_lengths),
+                                                static_cast<std::size_t>(1),
+                                                std::multiplies<>());
+}
+
+template <typename WeiDataType, ck::index_t NumDimSpatial, ck::index_t NumNonSpatialDim = 3>
+std::size_t
+GetWeightByte(const std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>& weights_lengths)
+{
+    // sizeof(WeiDataType) * (G * K * C * <filter spatial lengths product>) +
+    return sizeof(WeiDataType) * std::accumulate(std::begin(weights_lengths),
+                                                 std::end(weights_lengths),
+                                                 static_cast<std::size_t>(1),
+                                                 std::multiplies<>());
+}
+
+template <typename OutDataType, ck::index_t NumDimSpatial, ck::index_t NumNonSpatialDim = 3>
+std::size_t
+GetOutputByte(const std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>& output_lengths)
+{
+    // sizeof(OutDataType) * (G * N * K * <output spatial lengths product>);
+    return sizeof(OutDataType) * std::accumulate(std::begin(output_lengths),
+                                                 std::end(output_lengths),
+                                                 static_cast<std::size_t>(1),
+                                                 std::multiplies<std::size_t>());
+}
+
+template <ck::index_t NumDimSpatial,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout,
+          ck::index_t NumNonSpatialDim = 3>
+bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> in_lengths,
+                          std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> wei_lengths,
+                          std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> out_lengths)
+{
+    std::size_t in_mem_size  = GetInputByte<InDataType, NumDimSpatial>(in_lengths);
+    std::size_t wei_mem_size = GetWeightByte<WeiDataType, NumDimSpatial>(wei_lengths);
+    std::size_t out_mem_size = GetOutputByte<OutDataType, NumDimSpatial>(out_lengths);
+
+    SimpleDeviceMem in(in_mem_size);
+    SimpleDeviceMem wei(wei_mem_size);
+    SimpleDeviceMem out(out_mem_size);
+
+    std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> in_strides;
+    std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> wei_strides;
+    std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> out_strides;
+    in_strides.fill(0);
+    wei_strides.fill(0);
+    out_strides.fill(0);
+    in_strides.back()  = 1;
+    wei_strides.back() = 1;
+    out_strides.back() = 1;
+
+    std::partial_sum(rbegin(in_lengths),
+                     std::prev(rend(in_lengths)),
+                     std::next(rbegin(in_strides)),
+                     std::multiplies<>{});
+    std::partial_sum(rbegin(wei_lengths),
+                     std::prev(rend(wei_lengths)),
+                     std::next(rbegin(wei_strides)),
+                     std::multiplies<>{});
+    std::partial_sum(rbegin(out_lengths),
+                     std::prev(rend(out_lengths)),
+                     std::next(rbegin(out_strides)),
+                     std::multiplies<>{});
+
+    // transpose NDHWGC/KZYXGC/NDHWGK to GNDHWC/GKZYXC/GNDHWK to GNCDHW/GKCZYX/GNKDHW
+    std::rotate(std::next(rbegin(in_lengths)), std::next(rbegin(in_lengths), 2), rend(in_lengths));
+    std::rotate(rbegin(in_lengths),
+                std::next(rbegin(in_lengths)),
+                std::next(rbegin(in_lengths), NumDimSpatial + 1));
+
+    std::rotate(std::next(rbegin(in_strides)), std::next(rbegin(in_strides), 2), rend(in_strides));
+    std::rotate(rbegin(in_strides),
+                std::next(rbegin(in_strides)),
+                std::next(rbegin(in_strides), NumDimSpatial + 1));
+
+    std::rotate(rbegin(wei_lengths),
+                std::next(rbegin(wei_lengths)),
+                std::next(rbegin(wei_lengths), NumDimSpatial + 1));
+
+    std::rotate(rbegin(wei_strides),
+                std::next(rbegin(wei_strides)),
+                std::next(rbegin(wei_strides), NumDimSpatial + 1));
+
+    std::rotate(
+        std::next(rbegin(out_lengths)), std::next(rbegin(out_lengths), 2), rend(out_lengths));
+    std::rotate(rbegin(out_lengths),
+                std::next(rbegin(out_lengths)),
+                std::next(rbegin(out_lengths), NumDimSpatial + 1));
+
+    std::rotate(
+        std::next(rbegin(out_strides)), std::next(rbegin(out_strides), 2), rend(out_strides));
+    std::rotate(rbegin(out_strides),
+                std::next(rbegin(out_strides)),
+                std::next(rbegin(out_strides), NumDimSpatial + 1));
+
+    std::array<ck::index_t, NumDimSpatial> conv_filter_strides;
+    std::array<ck::index_t, NumDimSpatial> conv_filter_dilations;
+    std::array<ck::index_t, NumDimSpatial> input_left_pads;
+    std::array<ck::index_t, NumDimSpatial> input_right_pads;
+    conv_filter_strides.fill(1);
+    conv_filter_dilations.fill(1);
+    input_left_pads.fill(1);
+    input_right_pads.fill(1);
+
+    std::size_t flop      = GetFlops<NumDimSpatial>(out_lengths, wei_lengths);
+    std::size_t num_bytes = in_mem_size + wei_mem_size + out_mem_size;
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
+                                                                                 InLayout,
+                                                                                 WeiLayout,
+                                                                                 ck::Tuple<>,
+                                                                                 OutLayout,
+                                                                                 InDataType,
+                                                                                 WeiDataType,
+                                                                                 ck::Tuple<>,
+                                                                                 OutDataType,
+                                                                                 PassThrough,
+                                                                                 PassThrough,
+                                                                                 PassThrough>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            in.GetDeviceBuffer(),
+            wei.GetDeviceBuffer(),
+            std::array<const void*, 0>{},
+            out.GetDeviceBuffer(),
+            in_lengths,
+            in_strides,
+            wei_lengths,
+            wei_strides,
+            std::array<std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>, 0>{{}},
+            std::array<std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>, 0>{{}},
+            out_lengths,
+            out_strides,
+            conv_filter_strides,
+            conv_filter_dilations,
+            input_left_pads,
+            input_right_pads,
+            PassThrough{},
+            PassThrough{},
+            PassThrough{});
+
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return false;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            in.GetDeviceBuffer(),
+            wei.GetDeviceBuffer(),
+            std::array<const void*, 0>{},
+            out.GetDeviceBuffer(),
+            in_lengths,
+            in_strides,
+            wei_lengths,
+            wei_strides,
+            std::array<std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>, 0>{{}},
+            std::array<std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim>, 0>{{}},
+            out_lengths,
+            out_strides,
+            conv_filter_strides,
+            conv_filter_dilations,
+            input_left_pads,
+            input_right_pads,
+            PassThrough{},
+            PassThrough{},
+            PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+    return true;
+}
diff --git a/client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp b/client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d4455df628afcb6ec721b5f884b79e4572a5037c
--- /dev/null
+++ b/client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::NDHWGK;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 64;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 3;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 3;
+
+int main()
+{
+    return run_grouped_conv_fwd<NumDimSpatial,
+                                InDataType,
+                                WeiDataType,
+                                OutDataType,
+                                InLayout,
+                                WeiLayout,
+                                OutLayout>(
+               {N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp b/client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..7e8c98b6037418001571707a82aa14987059fb4f
--- /dev/null
+++ b/client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = float;
+using WeiDataType = float;
+using OutDataType = float;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::NDHWGK;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 64;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 3;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 3;
+
+int main()
+{
+    return run_grouped_conv_fwd<NumDimSpatial,
+                                InDataType,
+                                WeiDataType,
+                                OutDataType,
+                                InLayout,
+                                WeiLayout,
+                                OutLayout>(
+               {N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt b/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..659e6769d8e69f227ea1ef5920c3951bba706f0d
--- /dev/null
+++ b/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
+target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_operations)
\ No newline at end of file
diff --git a/client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp b/client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..7ba3224fc3244ab3bd472b98864efa42ae132fde
--- /dev/null
+++ b/client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp
@@ -0,0 +1,232 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+#include <random>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_fastgelu.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using FastGelu    = ck::tensor_operation::element_wise::FastGelu;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using DsDataType = ck::Tuple<>;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = FastGelu;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::mt19937 gen(19391);
+    std::uniform_int_distribution<> distrib(1, 10);
+    int group_count = distrib(gen);
+
+    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideEs;
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        Ms.push_back(256 + 256 * distrib(gen));
+        Ns.push_back(256 + 256 * distrib(gen));
+        Ks.push_back(128 + 128 * distrib(gen));
+
+        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
+        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
+        StrideEs.push_back(std::is_same<Row, ELayout>::value ? Ns[i] : Ms[i]);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    std::vector<SimpleDeviceMem> a_dev_bufs, b_dev_bufs, e_dev_bufs;
+
+    a_dev_bufs.reserve(group_count);
+    b_dev_bufs.reserve(group_count);
+    e_dev_bufs.reserve(group_count);
+
+    std::vector<const void*> p_a, p_b;
+    std::vector<void*> p_e;
+
+    p_a.reserve(group_count);
+    p_b.reserve(group_count);
+    p_e.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+
+    gemm_descs.reserve(group_count);
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        a_dev_bufs.emplace_back(sizeof(ADataType) *
+                                f_matrix_space_size(Ms[i], Ks[i], StrideAs[i], ALayout{}));
+        b_dev_bufs.emplace_back(sizeof(BDataType) *
+                                f_matrix_space_size(Ks[i], Ns[i], StrideBs[i], BLayout{}));
+        e_dev_bufs.emplace_back(sizeof(EDataType) *
+                                f_matrix_space_size(Ms[i], Ns[i], StrideEs[i], ELayout{}));
+
+        gemm_descs.push_back({Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideEs[i], {}});
+
+        p_a.push_back(a_dev_bufs[i].GetDeviceBuffer());
+        p_b.push_back(b_dev_bufs[i].GetDeviceBuffer());
+        p_e.push_back(e_dev_bufs[i].GetDeviceBuffer());
+    }
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedGemm<ALayout,
+                                                                     BLayout,
+                                                                     DsLayout,
+                                                                     ELayout,
+                                                                     ADataType,
+                                                                     BDataType,
+                                                                     DsDataType,
+                                                                     EDataType,
+                                                                     AElementOp,
+                                                                     BElementOp,
+                                                                     CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    auto p_ds = std::vector<std::array<const void*, 0>>{};
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            p_a, p_b, p_ds, p_e, gemm_descs, a_element_op, b_element_op, cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        SimpleDeviceMem gemm_desc_workspace(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(), gemm_desc_workspace.GetDeviceBuffer());
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = 0, num_btype = 0;
+            for(std::size_t j = 0; j < gemm_descs.size(); ++j)
+            {
+                flop += std::size_t(2) * Ms[j] * Ns[j] * Ks[j];
+
+                num_btype += sizeof(ADataType) * Ms[j] * Ks[j] + sizeof(BDataType) * Ks[j] * Ns[j] +
+                             sizeof(EDataType) * Ms[j] * Ns[j];
+            }
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    if(found)
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            p_a, p_b, p_ds, p_e, gemm_descs, a_element_op, b_element_op, cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        SimpleDeviceMem gemm_desc_workspace(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(), gemm_desc_workspace.GetDeviceBuffer());
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/18_groupnorm/CMakeLists.txt b/client_example/18_groupnorm/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..17c88cb61bc5daaa3ac0c6a92459c147691be2fe
--- /dev/null
+++ b/client_example/18_groupnorm/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_executable(client_groupnorm_swish groupnorm_swish.cpp)
+target_link_libraries(client_groupnorm_swish PRIVATE composable_kernel::device_operations)
diff --git a/client_example/18_groupnorm/groupnorm_swish.cpp b/client_example/18_groupnorm/groupnorm_swish.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e1d198d2282b85f21fb4c04afde443133c801b15
--- /dev/null
+++ b/client_example/18_groupnorm/groupnorm_swish.cpp
@@ -0,0 +1,194 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_normalization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/normalization_swish.hpp"
+
+using XDataType       = ck::half_t;
+using GammaDataType   = float;
+using BetaDataType    = float;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using Swish           = ck::tensor_operation::element_wise::Swish;
+
+constexpr int Rank         = 5;
+constexpr int NumReduceDim = 3;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t N = 32;
+    ck::index_t H = 16;
+    ck::index_t W = 16;
+    ck::index_t G = 64;
+    ck::index_t C = 128;
+
+    std::size_t xy_size         = N * H * W * G * C;
+    std::size_t gamma_beta_size = G * C;
+
+    std::vector<ck::index_t> xy_strides         = {H * W * G * C, W * G * C, G * C, C, 1};
+    std::vector<ck::index_t> gamma_beta_strides = {0, 0, 0, C, 1};
+
+    SimpleDeviceMem x_device_buf(sizeof(XDataType) * xy_size);
+    SimpleDeviceMem gamma_device_buf(sizeof(GammaDataType) * gamma_beta_size);
+    SimpleDeviceMem beta_device_buf(sizeof(BetaDataType) * gamma_beta_size);
+    SimpleDeviceMem y_device_buf(sizeof(YDataType) * xy_size);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
+                                                                       GammaDataType,
+                                                                       BetaDataType,
+                                                                       ComputeDataType,
+                                                                       YDataType,
+                                                                       Swish,
+                                                                       Rank,
+                                                                       NumReduceDim>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto& generic_op_ptr = op_ptrs[0];
+
+    auto generic_argument_ptr =
+        generic_op_ptr->MakeArgumentPointer({N, H, W, G, C},    // lengths
+                                            xy_strides,         // xStrides
+                                            gamma_beta_strides, // gammaStrides
+                                            gamma_beta_strides, // betaStrides
+                                            xy_strides,         // yStrides
+                                            {1, 2, 4},          // reduceDims
+                                            1e-6,
+                                            x_device_buf.GetDeviceBuffer(),
+                                            gamma_device_buf.GetDeviceBuffer(),
+                                            beta_device_buf.GetDeviceBuffer(),
+                                            y_device_buf.GetDeviceBuffer(),
+                                            nullptr,
+                                            nullptr,
+                                            Swish{});
+
+    if(!generic_op_ptr->IsSupportedArgument(generic_argument_ptr.get()))
+    {
+        throw std::runtime_error(
+            "The generic kernel instance should be able to support any input shapes");
+    };
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},    // lengths
+                                                        xy_strides,         // xStrides
+                                                        gamma_beta_strides, // gammaStrides
+                                                        gamma_beta_strides, // betaStrides
+                                                        xy_strides,         // yStrides
+                                                        {1, 2, 4},          // reduceDims
+                                                        1e-6,
+                                                        x_device_buf.GetDeviceBuffer(),
+                                                        gamma_device_buf.GetDeviceBuffer(),
+                                                        beta_device_buf.GetDeviceBuffer(),
+                                                        y_device_buf.GetDeviceBuffer(),
+                                                        nullptr,
+                                                        nullptr,
+                                                        Swish{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_byte =
+                sizeof(XDataType) * xy_size + sizeof(GammaDataType) * gamma_beta_size +
+                sizeof(BetaDataType) * gamma_beta_size + sizeof(YDataType) * xy_size;
+
+            float gb_per_sec = num_byte / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},    // lengths
+                                                        xy_strides,         // xStrides
+                                                        gamma_beta_strides, // gammaStrides
+                                                        gamma_beta_strides, // betaStrides
+                                                        xy_strides,         // yStrides
+                                                        {1, 2, 4},          // reduceDims
+                                                        1e-6,
+                                                        x_device_buf.GetDeviceBuffer(),
+                                                        gamma_device_buf.GetDeviceBuffer(),
+                                                        beta_device_buf.GetDeviceBuffer(),
+                                                        y_device_buf.GetDeviceBuffer(),
+                                                        nullptr,
+                                                        nullptr,
+                                                        Swish{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/19_pool_fwd/CMakeLists.txt b/client_example/19_pool_fwd/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..13f9f73c83d55c801fdf4609e13f6b0813cb0c67
--- /dev/null
+++ b/client_example/19_pool_fwd/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_executable(client_max_pool2d_fwd max_pool2d_fwd.cpp)
+target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_operations)
+
+add_executable(client_avg_pool3d_fwd avg_pool3d_fwd.cpp)
+target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_operations)
\ No newline at end of file
diff --git a/client_example/19_pool_fwd/avg_pool3d_fwd.cpp b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..2edaf474b5644a0a91fb51e675e1c37286841967
--- /dev/null
+++ b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
@@ -0,0 +1,199 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
+
+using InDataType    = ck::half_t;
+using OutDataType   = ck::half_t;
+using IndexDataType = int32_t;
+
+constexpr ck::index_t InOutRank  = 5;
+constexpr ck::index_t WindowRank = 3;
+#if 0
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::MAX;
+constexpr bool OutputIndex = false;
+#else
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::AVG;
+constexpr bool OutputIndex = false;
+#endif
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t N               = 2;
+    ck::index_t C               = 32;
+    ck::index_t Z               = 2;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Di              = 30;
+    ck::index_t Hi              = 30;
+    ck::index_t Wi              = 30;
+    ck::index_t window_stride_d = 2;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_d   = 1;
+    ck::index_t in_left_pad_h   = 1;
+    ck::index_t in_left_pad_w   = 1;
+    ck::index_t in_right_pad_d  = 1;
+    ck::index_t in_right_pad_h  = 1;
+    ck::index_t in_right_pad_w  = 1;
+
+    ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Z) / window_stride_d + 1;
+    ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    // Pool API only support the order of NCDHW
+    std::vector<ck::index_t> in_length              = {N, C, Di, Hi, Wi};
+    std::vector<ck::index_t> out_length             = {N, C, Do, Ho, Wo};
+    std::vector<ck::index_t> window_spatial_lengths = {Z, Y, X};
+    std::vector<ck::index_t> window_strides   = {window_stride_d, window_stride_h, window_stride_w};
+    std::vector<ck::index_t> input_left_pads  = {in_left_pad_d, in_left_pad_h, in_left_pad_w};
+    std::vector<ck::index_t> input_right_pads = {in_right_pad_d, in_right_pad_h, in_right_pad_w};
+
+    std::size_t in_tensor_size  = N * C * Di * Hi * Wi;
+    std::size_t out_tensor_size = N * C * Do * Ho * Wo;
+
+    // tensor layout = NDHWC
+    std::vector<ck::index_t> in_tensor_stride  = {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C};
+    std::vector<ck::index_t> out_tensor_stride = {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C};
+
+    SimpleDeviceMem in_device_buf(sizeof(InDataType) * in_tensor_size);
+    SimpleDeviceMem out_device_buf(sizeof(OutDataType) * out_tensor_size);
+    SimpleDeviceMem out_indices_device_buf(sizeof(IndexDataType) * out_tensor_size);
+
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3, 4});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes =
+                in_tensor_size * sizeof(InDataType) + out_tensor_size * sizeof(OutDataType);
+
+            if constexpr(OutputIndex)
+                num_bytes += out_tensor_size * sizeof(IndexDataType);
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3, 4});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/19_pool_fwd/max_pool2d_fwd.cpp b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c776dc12da391de21daa2cf464d2bde39873db9f
--- /dev/null
+++ b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
@@ -0,0 +1,193 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp"
+
+using InDataType    = ck::half_t;
+using OutDataType   = ck::half_t;
+using IndexDataType = int32_t;
+
+constexpr ck::index_t InOutRank  = 4;
+constexpr ck::index_t WindowRank = 2;
+#if 1
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::MAX;
+constexpr bool OutputIndex = true;
+#else
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::AVG;
+constexpr bool OutputIndex = false;
+#endif
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t N               = 2;
+    ck::index_t C               = 32;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Hi              = 30;
+    ck::index_t Wi              = 30;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_h   = 1;
+    ck::index_t in_left_pad_w   = 1;
+    ck::index_t in_right_pad_h  = 1;
+    ck::index_t in_right_pad_w  = 1;
+
+    ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    // Pool API only support the order of NCHW
+    std::vector<ck::index_t> in_length              = {N, C, Hi, Wi};
+    std::vector<ck::index_t> out_length             = {N, C, Ho, Wo};
+    std::vector<ck::index_t> window_spatial_lengths = {Y, X};
+    std::vector<ck::index_t> window_strides         = {window_stride_h, window_stride_w};
+    std::vector<ck::index_t> input_left_pads        = {in_left_pad_h, in_left_pad_w};
+    std::vector<ck::index_t> input_right_pads       = {in_right_pad_h, in_right_pad_w};
+
+    std::size_t in_tensor_size  = N * C * Hi * Wi;
+    std::size_t out_tensor_size = N * C * Ho * Wo;
+
+    // tensor layout = NHWC
+    std::vector<ck::index_t> in_tensor_stride  = {C * Hi * Wi, 1, Wi * C, C};
+    std::vector<ck::index_t> out_tensor_stride = {C * Ho * Wo, 1, Wo * C, C};
+
+    SimpleDeviceMem in_device_buf(sizeof(InDataType) * in_tensor_size);
+    SimpleDeviceMem out_device_buf(sizeof(OutDataType) * out_tensor_size);
+    SimpleDeviceMem out_indices_device_buf(sizeof(IndexDataType) * out_tensor_size);
+
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes =
+                in_tensor_size * sizeof(InDataType) + out_tensor_size * sizeof(OutDataType);
+
+            if constexpr(OutputIndex)
+                num_bytes += out_tensor_size * sizeof(IndexDataType);
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/cmake/EnableCompilerWarnings.cmake b/cmake/EnableCompilerWarnings.cmake
index 78133af0315dd952771ecb481525c14f8f4abff5..369cd0b54c1fac60b061cdafdb0b02b819a8accd 100644
--- a/cmake/EnableCompilerWarnings.cmake
+++ b/cmake/EnableCompilerWarnings.cmake
@@ -65,7 +65,8 @@ else()
             -Wuninitialized
             -Wunreachable-code
             -Wunused
-
+            -Wno-reserved-identifier
+            -Werror
             -Wsign-compare
             -Wno-extra-semi-stmt
         )
@@ -91,6 +92,7 @@ else()
                 -Wno-unused-command-line-argument
                 -Wno-weak-vtables
                 -Wno-covered-switch-default
+                -Wno-unsafe-buffer-usage
             )
         else()
             if (CMAKE_${COMPILER}_COMPILER_ID MATCHES "GNU" AND ${COMPILER} MATCHES "CXX")
diff --git a/cmake/googletest.cmake b/cmake/googletest.cmake
index 3c6cb56ccea3490c6d7a4eaf6107749dafa87bdf..d6577ac33e71c575b2aa441da30838f58157f0b4 100644
--- a/cmake/googletest.cmake
+++ b/cmake/googletest.cmake
@@ -21,6 +21,7 @@ list(APPEND GTEST_CMAKE_CXX_FLAGS
      -Wno-comma
      -Wno-old-style-cast
      -Wno-deprecated
+     -Wno-unsafe-buffer-usage
 )
 message(STATUS "Suppressing googltest warnings with flags: ${GTEST_CMAKE_CXX_FLAGS}")
 
diff --git a/doc/markdown/dockerhub.md b/doc/markdown/dockerhub.md
deleted file mode 100644
index 91b6cb2295cf4a8dd08ead2eb165119c74996084..0000000000000000000000000000000000000000
--- a/doc/markdown/dockerhub.md
+++ /dev/null
@@ -1,93 +0,0 @@
-## CK docker hub
-
-[Docker hub](https://hub.docker.com/r/rocm/composable_kernel)
-
-## Why do I need this?
-
-To make our lives easier and bring Composable Kernel dependencies together, we recommend using docker images.
-
-## So what is Composable Kernel?
-
-Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
-
-To get the CK library
-
-```
-git clone https://github.com/ROCmSoftwarePlatform/composable_kernel.git
-```
-
-run a docker container 
-
-```
-docker run                                                            \
--it                                                                   \
---privileged                                                          \
---group-add sudo                                                      \
--w /root/workspace                                                    \
--v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace                         \
-rocm/composable_kernel:ck_ub20.04_rocm5.3_release                     \
-/bin/bash
-```
-
-and build the CK
-
-```
-mkdir build && cd build
-
-# Need to specify target ID, example below is for gfx908 and gfx90a
-cmake                                                                                             \
--D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
--D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
--D CMAKE_CXX_FLAGS="-O3"                                                                          \
--D CMAKE_BUILD_TYPE=Release                                                                       \
--D GPU_TARGETS="gfx908;gfx90a"                                                                    \
-..
-```
-
-and 
-
-```
-make -j examples tests
-```
-
-To run all the test cases including tests and examples run
-
-```
-make test
-```
-
-We can also run specific examples or tests like
-
-```
-./bin/example_gemm_xdl_fp16
-./bin/test_gemm_fp16
-```
-
-For more details visit [CK github repo](https://github.com/ROCmSoftwarePlatform/composable_kernel), [CK examples](https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/example), [even more CK examples](https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/client_example).
-
-## And what is inside?
-
-The docker images have everything you need for running CK including:
-
-* [ROCm](https://www.amd.com/en/graphics/servers-solutions-rocm)
-* [CMake](https://cmake.org/)
-* [Compiler](https://github.com/RadeonOpenCompute/llvm-project)
-
-## Which image is right for me?
-
-Let's take a look at the image naming, for example "ck_ub20.04_rocm5.4_release". The image specs are:
-
-* "ck" - made for running Composable Kernel
-* "ub20.04" - based on Ubuntu 20.04
-* "rocm5.4" - ROCm platform version 5.4
-* "release" - compiler version is release
-
-So just pick the right image for your project dependencies and you're all set.
-
-## DIY starts here
-
-If you need to customize a docker image or just can't stop tinkering, feel free to adjust the [Dockerfile](https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/Dockerfile) for your needs.
-
-## License
-
-CK is released under the MIT [license](https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/LICENSE).
diff --git a/docs/API_Reference_Guide.rst b/docs/API_Reference_Guide.rst
new file mode 100644
index 0000000000000000000000000000000000000000..b59c6e302692cd6b9d4035ed27d5901e790a0233
--- /dev/null
+++ b/docs/API_Reference_Guide.rst
@@ -0,0 +1,52 @@
+
+*******************
+API Reference Guide
+*******************
+
+=================
+Introduction
+=================
+
+This document contains details of the APIs for the Composable Kernel (CK) library and introduces some of the key design
+principles that are used to write new classes that extend CK functionality.
+
+=================
+Using CK API
+=================
+
+This section describes how to use the CK library API.
+
+=================
+CK Datatypes
+=================
+
+-----------------
+DeviceMem
+-----------------
+
+.. doxygenstruct:: DeviceMem
+
+---------------------------
+Kernels For Flashattention
+---------------------------
+
+The Flashattention algorithm is defined in :cite:t:`dao2022flashattention`.  This sections lists the classes that are
+used in the CK GPU implementation of Flashattention.
+
+**Gridwise classes**
+
+.. doxygenstruct:: ck::GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
+
+**Blockwise classes**
+
+.. doxygenstruct:: ck::ThreadGroupTensorSliceTransfer_v4r1
+
+.. doxygenstruct:: ck::BlockwiseGemmXdlops_v2
+
+.. doxygenstruct:: ck::BlockwiseSoftmax
+
+**Threadwise classes**
+
+.. doxygenstruct:: ck::ThreadwiseTensorSliceTransfer_StaticToStatic
+
+.. bibliography::
diff --git a/docs/Contributors_Guide.rst b/docs/Contributors_Guide.rst
new file mode 100644
index 0000000000000000000000000000000000000000..b2ddff398ce8efcd0c5d8be275d9dd09315d4349
--- /dev/null
+++ b/docs/Contributors_Guide.rst
@@ -0,0 +1,8 @@
+===================
+Contributor's Guide
+===================
+
+Pull-request guidelines
+=======================
+
+[TODO]
diff --git a/docs/Supported_Primitives_Guide.rst b/docs/Supported_Primitives_Guide.rst
new file mode 100644
index 0000000000000000000000000000000000000000..4c3adf67d7119e22a622373ab8d1dccb4d024e18
--- /dev/null
+++ b/docs/Supported_Primitives_Guide.rst
@@ -0,0 +1,75 @@
+==========================
+Supported Primitives Guide
+==========================
+
+This document contains details of supported primitives in Composable Kernel (CK). In contrast to the API Reference
+Guide, the Supported Primitives Guide is an introduction to the math which underpins the algorithms implemented in CK.
+
+------------
+Softmax
+------------
+
+For vectors :math:`x^{(1)}, x^{(2)}, \ldots, x^{(T)}` of size :math:`B` we can decompose the softmax of concatenated
+:math:`x = [ x^{(1)}\ | \ \ldots \ | \ x^{(T)} ]` as,
+
+.. math::
+   :nowrap:
+
+   \begin{align}
+      m(x) & = m( [ x^{(1)}\ | \ \ldots \ | \ x^{(T)} ] ) = \max( m(x^{(1)}),\ldots, m(x^{(T)}) )  \\
+      f(x) & = [\exp( m(x^{(1)}) - m(x) ) f( x^{(1)} )\ | \ \ldots \ | \ \exp( m(x^{(T)}) - m(x) ) f( x^{(T)} )] \\
+      z(x) & = \exp( m(x^{(1)}) - m(x) )\ z(x^{(1)}) + \ldots + \exp( m(x^{(T)}) - m(x) )\ z(x^{(1)}) \\
+      \operatorname{softmax}(x) &= f(x)\ / \ z(x)
+   \end{align}
+
+where :math:`f(x^{(j)}) = \exp( x^{(j)} - m(x^{(j)}) )` is of size :math:`B` and
+:math:`z(x^{(j)}) = f(x_1^{(j)})+ \ldots+ f(x_B^{(j)})` is a scalar.
+
+For a matrix :math:`X` composed of :math:`T_r \times T_c` tiles, :math:`X_{ij}`, of size :math:`B_r \times B_c` we can
+compute the row-wise softmax as follows.
+
+For :math:`j` from :math:`1` to :math:`T_c`, and :math:`i` from :math:`1` to :math:`T_r` calculate,
+
+.. math::
+   :nowrap:
+
+   \begin{align}
+      \tilde{m}_{ij}   &= \operatorname{rowmax}( X_{ij} ) \\
+      \tilde{P}_{ij}   &= \exp(X_{ij} - \tilde{m}_{ij} ) \\
+      \tilde{z}_{ij}   &= \operatorname{rowsum}( P_{ij} ) \\
+   \end{align}
+
+If :math:`j=1`, initialize running max, running sum, and the first column block of the output,
+
+.. math::
+   :nowrap:
+
+   \begin{align}
+      m_i            &= \tilde{m}_{i1} \\
+      z_i            &= \tilde{z}_{i1} \\
+      \tilde{Y}_{i1} &= \diag(\tilde{z}_{ij})^{-1} \tilde{P}_{i1}
+   \end{align}
+
+Else if :math:`j>1`,
+
+1. Update running max, running sum and column blocks :math:`k=1` to :math:`k=j-1`
+
+.. math::
+   :nowrap:
+
+   \begin{align}
+      m^{new}_i &= \max(m_i, \tilde{m}_{ij} ) \\
+      z^{new}_i &= \exp(m_i - m^{new}_i)\ z_i + \exp( \tilde{m}_{ij} - m^{new}_i )\ \tilde{z}_{ij}  \\
+      Y_{ik}    &= \diag(z^{new}_{i})^{-1} \diag(z_{i}) \exp(m_i - m^{new}_i)\ Y_{ik}
+   \end{align}
+
+2. Initialize column block :math:`j` of output and reset running max and running sum variables:
+
+.. math::
+   :nowrap:
+
+   \begin{align}
+      \tilde{Y}_{ij} &= \diag(z^{new}_{i})^{-1} \exp(\tilde{m}_{ij} - m^{new}_i ) \tilde{P}_{ij} \\
+      z_i            &= z^{new}_i \\
+      m_i            &= m^{new}_i \\
+   \end{align}
diff --git a/docs/conf.py b/docs/conf.py
new file mode 100644
index 0000000000000000000000000000000000000000..0de590da1a53daba8dc0e379e68abbae250e0667
--- /dev/null
+++ b/docs/conf.py
@@ -0,0 +1,36 @@
+# Configuration file for the Sphinx documentation builder.
+#
+# This file only contains a selection of the most common options. For a full
+# list see the documentation:
+# https://www.sphinx-doc.org/en/master/usage/configuration.html
+
+import subprocess
+
+from rocm_docs import ROCmDocs
+
+
+name = "Composable Kernel"
+get_version = r'sed -n -e "s/^rocm_setup_version(.* \([0-9\.]\{1,\}\).*/\1/p" ../CMakeLists.txt'
+version = subprocess.getoutput(get_version)
+if len(version) > 0:
+    name = f"{name} {version}"
+
+external_toc_path = "./sphinx/_toc.yml"
+
+docs_core = ROCmDocs(f"{name} Documentation")
+docs_core.run_doxygen(doxygen_root="doxygen", doxygen_path="doxygen/docBin/xml")
+docs_core.setup()
+
+mathjax3_config = {
+'tex': {
+    'macros': {
+        'diag': '\\operatorname{diag}',
+        }
+    }
+}
+
+for sphinx_var in ROCmDocs.SPHINX_VARS:
+    globals()[sphinx_var] = getattr(docs_core, sphinx_var)
+
+extensions += ['sphinxcontrib.bibtex']
+bibtex_bibfiles = ['refs.bib']
diff --git a/doc/image/ck_component.png b/docs/data/ck_component.png
similarity index 100%
rename from doc/image/ck_component.png
rename to docs/data/ck_component.png
diff --git a/doc/image/ck_layer.png b/docs/data/ck_layer.png
similarity index 100%
rename from doc/image/ck_layer.png
rename to docs/data/ck_layer.png
diff --git a/docs/dockerhub.rst b/docs/dockerhub.rst
new file mode 100644
index 0000000000000000000000000000000000000000..b51226cfebe6da27988ec2685eb334f5bc900ff9
--- /dev/null
+++ b/docs/dockerhub.rst
@@ -0,0 +1,96 @@
+===================
+CK docker hub
+===================
+
+`Docker hub <https://hub.docker.com/r/rocm/composable_kernel>`_
+
+-------------------------------------
+Why do I need this?
+-------------------------------------
+
+To make our lives easier and bring Composable Kernel dependencies together, we recommend using docker images.
+
+-------------------------------------
+So what is Composable Kernel?
+-------------------------------------
+
+Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
+
+To get the CK library::
+
+    git clone https://github.com/ROCmSoftwarePlatform/composable_kernel.git
+
+
+
+run a docker container::
+
+    docker run                                                            \
+    -it                                                                   \
+    --privileged                                                          \
+    --group-add sudo                                                      \
+    -w /root/workspace                                                    \
+    -v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace                         \
+    rocm/composable_kernel:ck_ub20.04_rocm5.3_release                     \
+    /bin/bash
+
+and build the CK::
+
+    mkdir build && cd build
+    # Need to specify target ID, example below is for gfx908 and gfx90a
+    cmake                                                                                             \
+    -D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
+    -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
+    -D CMAKE_CXX_FLAGS="-O3"                                                                          \
+    -D CMAKE_BUILD_TYPE=Release                                                                       \
+    -D GPU_TARGETS="gfx908;gfx90a"                                                                    \
+    ..
+
+and::
+
+    make -j examples tests
+
+To run all the test cases including tests and examples run::
+
+    make test
+
+We can also run specific examples or tests like::
+
+    ./bin/example_gemm_xdl_fp16
+    ./bin/test_gemm_fp16
+
+For more details visit `CK github repo <https://github.com/ROCmSoftwarePlatform/composable_kernel>`_, `CK examples <https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/example)>`_, `even more CK examples <https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/client_example>`_.
+
+-------------------------------------
+And what is inside?
+-------------------------------------
+
+The docker images have everything you need for running CK including:
+
+* `ROCm <https://www.amd.com/en/graphics/servers-solutions-rocm>`_
+* `CMake <https://cmake.org/>`_
+* `Compiler <https://github.com/RadeonOpenCompute/llvm-project>`_
+
+-------------------------------------
+Which image is right for me?
+-------------------------------------
+
+Let's take a look at the image naming, for example "ck_ub20.04_rocm5.4_release". The image specs are:
+
+* "ck" - made for running Composable Kernel
+* "ub20.04" - based on Ubuntu 20.04
+* "rocm5.4" - ROCm platform version 5.4
+* "release" - compiler version is release
+
+So just pick the right image for your project dependencies and you're all set.
+
+-------------------------------------
+DIY starts here
+-------------------------------------
+
+If you need to customize a docker image or just can't stop tinkering, feel free to adjust the `Dockerfile <https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/Dockerfile>`_ for your needs.
+
+-------------------------------------
+License
+-------------------------------------
+
+CK is released under the MIT `license <https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/LICENSE>`_.
diff --git a/docs/doxygen/Doxyfile b/docs/doxygen/Doxyfile
new file mode 100644
index 0000000000000000000000000000000000000000..1084f94c81bbef88a3373d6e9e5d24b89e7b5353
--- /dev/null
+++ b/docs/doxygen/Doxyfile
@@ -0,0 +1,2455 @@
+# Doxyfile 1.8.10
+
+# This file describes the settings to be used by the documentation system
+# doxygen (www.doxygen.org) for a project.
+#
+# All text after a double hash (##) is considered a comment and is placed in
+# front of the TAG it is preceding.
+#
+# All text after a single hash (#) is considered a comment and will be ignored.
+# The format is:
+# TAG = value [value, ...]
+# For lists, items can also be appended using:
+# TAG += value [value, ...]
+# Values that contain spaces should be placed between quotes (\" \").
+
+#---------------------------------------------------------------------------
+# Project related configuration options
+#---------------------------------------------------------------------------
+
+# This tag specifies the encoding used for all characters in the config file
+# that follow. The default is UTF-8 which is also the encoding used for all text
+# before the first occurrence of this tag. Doxygen uses libiconv (or the iconv
+# built into libc) for the transcoding. See http://www.gnu.org/software/libiconv
+# for the list of possible encodings.
+# The default value is: UTF-8.
+
+DOXYFILE_ENCODING      = UTF-8
+
+# The PROJECT_NAME tag is a single word (or a sequence of words surrounded by
+# double-quotes, unless you are using Doxywizard) that should identify the
+# project for which the documentation is generated. This name is used in the
+# title of most generated pages and in a few other places.
+# The default value is: My Project.
+
+PROJECT_NAME           = "ck"
+
+# The PROJECT_NUMBER tag can be used to enter a project or revision number. This
+# could be handy for archiving the generated documentation or if some version
+# control system is used.
+
+PROJECT_NUMBER         = v3.0.1.0
+
+# Using the PROJECT_BRIEF tag one can provide an optional one line description
+# for a project that appears at the top of each page and should give viewer a
+# quick idea about the purpose of the project. Keep the description short.
+
+PROJECT_BRIEF          = "prototype interfaces compatible with ROCm platform and HiP"
+
+# With the PROJECT_LOGO tag one can specify a logo or an icon that is included
+# in the documentation. The maximum height of the logo should not exceed 55
+# pixels and the maximum width should not exceed 200 pixels. Doxygen will copy
+# the logo to the output directory.
+
+PROJECT_LOGO           = 
+
+# The OUTPUT_DIRECTORY tag is used to specify the (relative or absolute) path
+# into which the generated documentation will be written. If a relative path is
+# entered, it will be relative to the location where doxygen was started. If
+# left blank the current directory will be used.
+
+OUTPUT_DIRECTORY       = docBin
+
+# If the CREATE_SUBDIRS tag is set to YES then doxygen will create 4096 sub-
+# directories (in 2 levels) under the output directory of each output format and
+# will distribute the generated files over these directories. Enabling this
+# option can be useful when feeding doxygen a huge amount of source files, where
+# putting all generated files in the same directory would otherwise causes
+# performance problems for the file system.
+# The default value is: NO.
+
+CREATE_SUBDIRS         = NO
+
+# If the ALLOW_UNICODE_NAMES tag is set to YES, doxygen will allow non-ASCII
+# characters to appear in the names of generated files. If set to NO, non-ASCII
+# characters will be escaped, for example _xE3_x81_x84 will be used for Unicode
+# U+3044.
+# The default value is: NO.
+
+ALLOW_UNICODE_NAMES    = NO
+
+# The OUTPUT_LANGUAGE tag is used to specify the language in which all
+# documentation generated by doxygen is written. Doxygen will use this
+# information to generate all constant output in the proper language.
+# Possible values are: Afrikaans, Arabic, Armenian, Brazilian, Catalan, Chinese,
+# Chinese-Traditional, Croatian, Czech, Danish, Dutch, English (United States),
+# Esperanto, Farsi (Persian), Finnish, French, German, Greek, Hungarian,
+# Indonesian, Italian, Japanese, Japanese-en (Japanese with English messages),
+# Korean, Korean-en (Korean with English messages), Latvian, Lithuanian,
+# Macedonian, Norwegian, Persian (Farsi), Polish, Portuguese, Romanian, Russian,
+# Serbian, Serbian-Cyrillic, Slovak, Slovene, Spanish, Swedish, Turkish,
+# Ukrainian and Vietnamese.
+# The default value is: English.
+
+OUTPUT_LANGUAGE        = English
+
+# If the BRIEF_MEMBER_DESC tag is set to YES, doxygen will include brief member
+# descriptions after the members that are listed in the file and class
+# documentation (similar to Javadoc). Set to NO to disable this.
+# The default value is: YES.
+
+BRIEF_MEMBER_DESC      = YES
+
+# If the REPEAT_BRIEF tag is set to YES, doxygen will prepend the brief
+# description of a member or function before the detailed description
+#
+# Note: If both HIDE_UNDOC_MEMBERS and BRIEF_MEMBER_DESC are set to NO, the
+# brief descriptions will be completely suppressed.
+# The default value is: YES.
+
+REPEAT_BRIEF           = YES
+
+# This tag implements a quasi-intelligent brief description abbreviator that is
+# used to form the text in various listings. Each string in this list, if found
+# as the leading text of the brief description, will be stripped from the text
+# and the result, after processing the whole list, is used as the annotated
+# text. Otherwise, the brief description is used as-is. If left blank, the
+# following values are used ($name is automatically replaced with the name of
+# the entity):The $name class, The $name widget, The $name file, is, provides,
+# specifies, contains, represents, a, an and the.
+
+ABBREVIATE_BRIEF       = "The $name class" \
+                         "The $name widget" \
+                         "The $name file" \
+                         is \
+                         provides \
+                         specifies \
+                         contains \
+                         represents \
+                         a \
+                         an \
+                         the
+
+# If the ALWAYS_DETAILED_SEC and REPEAT_BRIEF tags are both set to YES then
+# doxygen will generate a detailed section even if there is only a brief
+# description.
+# The default value is: NO.
+
+ALWAYS_DETAILED_SEC    = NO
+
+# If the INLINE_INHERITED_MEMB tag is set to YES, doxygen will show all
+# inherited members of a class in the documentation of that class as if those
+# members were ordinary class members. Constructors, destructors and assignment
+# operators of the base classes will not be shown.
+# The default value is: NO.
+
+INLINE_INHERITED_MEMB  = NO
+
+# If the FULL_PATH_NAMES tag is set to YES, doxygen will prepend the full path
+# before files name in the file list and in the header files. If set to NO the
+# shortest path that makes the file name unique will be used
+# The default value is: YES.
+
+FULL_PATH_NAMES        = YES
+
+# The STRIP_FROM_PATH tag can be used to strip a user-defined part of the path.
+# Stripping is only done if one of the specified strings matches the left-hand
+# part of the path. The tag can be used to show relative paths in the file list.
+# If left blank the directory from which doxygen is run is used as the path to
+# strip.
+#
+# Note that you can specify absolute paths here, but also relative paths, which
+# will be relative from the directory where doxygen is started.
+# This tag requires that the tag FULL_PATH_NAMES is set to YES.
+
+STRIP_FROM_PATH        =
+
+# The STRIP_FROM_INC_PATH tag can be used to strip a user-defined part of the
+# path mentioned in the documentation of a class, which tells the reader which
+# header file to include in order to use a class. If left blank only the name of
+# the header file containing the class definition is used. Otherwise one should
+# specify the list of include paths that are normally passed to the compiler
+# using the -I flag.
+
+STRIP_FROM_INC_PATH    =
+
+# If the SHORT_NAMES tag is set to YES, doxygen will generate much shorter (but
+# less readable) file names. This can be useful is your file systems doesn't
+# support long names like on DOS, Mac, or CD-ROM.
+# The default value is: NO.
+
+SHORT_NAMES            = NO
+
+# If the JAVADOC_AUTOBRIEF tag is set to YES then doxygen will interpret the
+# first line (until the first dot) of a Javadoc-style comment as the brief
+# description. If set to NO, the Javadoc-style will behave just like regular Qt-
+# style comments (thus requiring an explicit @brief command for a brief
+# description.)
+# The default value is: NO.
+
+JAVADOC_AUTOBRIEF      = NO
+
+# If the QT_AUTOBRIEF tag is set to YES then doxygen will interpret the first
+# line (until the first dot) of a Qt-style comment as the brief description. If
+# set to NO, the Qt-style will behave just like regular Qt-style comments (thus
+# requiring an explicit \brief command for a brief description.)
+# The default value is: NO.
+
+QT_AUTOBRIEF           = NO
+
+# The MULTILINE_CPP_IS_BRIEF tag can be set to YES to make doxygen treat a
+# multi-line C++ special comment block (i.e. a block of //! or /// comments) as
+# a brief description. This used to be the default behavior. The new default is
+# to treat a multi-line C++ comment block as a detailed description. Set this
+# tag to YES if you prefer the old behavior instead.
+#
+# Note that setting this tag to YES also means that rational rose comments are
+# not recognized any more.
+# The default value is: NO.
+
+MULTILINE_CPP_IS_BRIEF = NO
+
+# If the INHERIT_DOCS tag is set to YES then an undocumented member inherits the
+# documentation from any documented member that it re-implements.
+# The default value is: YES.
+
+INHERIT_DOCS           = YES
+
+# If the SEPARATE_MEMBER_PAGES tag is set to YES then doxygen will produce a new
+# page for each member. If set to NO, the documentation of a member will be part
+# of the file/class/namespace that contains it.
+# The default value is: NO.
+
+SEPARATE_MEMBER_PAGES  = NO
+
+# The TAB_SIZE tag can be used to set the number of spaces in a tab. Doxygen
+# uses this value to replace tabs by spaces in code fragments.
+# Minimum value: 1, maximum value: 16, default value: 4.
+
+TAB_SIZE               = 4
+
+# This tag can be used to specify a number of aliases that act as commands in
+# the documentation. An alias has the form:
+# name=value
+# For example adding
+# "sideeffect=@par Side Effects:\n"
+# will allow you to put the command \sideeffect (or @sideeffect) in the
+# documentation, which will result in a user-defined paragraph with heading
+# "Side Effects:". You can put \n's in the value part of an alias to insert
+# newlines.
+
+ALIASES                =
+
+# This tag can be used to specify a number of word-keyword mappings (TCL only).
+# A mapping has the form "name=value". For example adding "class=itcl::class"
+# will allow you to use the command class in the itcl::class meaning.
+
+TCL_SUBST              =
+
+# Set the OPTIMIZE_OUTPUT_FOR_C tag to YES if your project consists of C sources
+# only. Doxygen will then generate output that is more tailored for C. For
+# instance, some of the names that are used will be different. The list of all
+# members will be omitted, etc.
+# The default value is: NO.
+
+OPTIMIZE_OUTPUT_FOR_C  = NO
+
+# Set the OPTIMIZE_OUTPUT_JAVA tag to YES if your project consists of Java or
+# Python sources only. Doxygen will then generate output that is more tailored
+# for that language. For instance, namespaces will be presented as packages,
+# qualified scopes will look different, etc.
+# The default value is: NO.
+
+OPTIMIZE_OUTPUT_JAVA   = NO
+
+# Set the OPTIMIZE_FOR_FORTRAN tag to YES if your project consists of Fortran
+# sources. Doxygen will then generate output that is tailored for Fortran.
+# The default value is: NO.
+
+OPTIMIZE_FOR_FORTRAN   = NO
+
+# Set the OPTIMIZE_OUTPUT_VHDL tag to YES if your project consists of VHDL
+# sources. Doxygen will then generate output that is tailored for VHDL.
+# The default value is: NO.
+
+OPTIMIZE_OUTPUT_VHDL   = NO
+
+# Doxygen selects the parser to use depending on the extension of the files it
+# parses. With this tag you can assign which parser to use for a given
+# extension. Doxygen has a built-in mapping, but you can override or extend it
+# using this tag. The format is ext=language, where ext is a file extension, and
+# language is one of the parsers supported by doxygen: IDL, Java, Javascript,
+# C#, C, C++, D, PHP, Objective-C, Python, Fortran (fixed format Fortran:
+# FortranFixed, free formatted Fortran: FortranFree, unknown formatted Fortran:
+# Fortran. In the later case the parser tries to guess whether the code is fixed
+# or free formatted code, this is the default for Fortran type files), VHDL. For
+# instance to make doxygen treat .inc files as Fortran files (default is PHP),
+# and .f files as C (default is Fortran), use: inc=Fortran f=C.
+#
+# Note: For files without extension you can use no_extension as a placeholder.
+#
+# Note that for custom extensions you also need to set FILE_PATTERNS otherwise
+# the files are not read by doxygen.
+
+EXTENSION_MAPPING      =
+
+# If the MARKDOWN_SUPPORT tag is enabled then doxygen pre-processes all comments
+# according to the Markdown format, which allows for more readable
+# documentation. See http://daringfireball.net/projects/markdown/ for details.
+# The output of markdown processing is further processed by doxygen, so you can
+# mix doxygen, HTML, and XML commands with Markdown formatting. Disable only in
+# case of backward compatibilities issues.
+# The default value is: YES.
+
+MARKDOWN_SUPPORT       = YES
+
+# When enabled doxygen tries to link words that correspond to documented
+# classes, or namespaces to their corresponding documentation. Such a link can
+# be prevented in individual cases by putting a % sign in front of the word or
+# globally by setting AUTOLINK_SUPPORT to NO.
+# The default value is: YES.
+
+AUTOLINK_SUPPORT       = YES
+
+# If you use STL classes (i.e. std::string, std::vector, etc.) but do not want
+# to include (a tag file for) the STL sources as input, then you should set this
+# tag to YES in order to let doxygen match functions declarations and
+# definitions whose arguments contain STL classes (e.g. func(std::string);
+# versus func(std::string) {}). This also make the inheritance and collaboration
+# diagrams that involve STL classes more complete and accurate.
+# The default value is: NO.
+
+BUILTIN_STL_SUPPORT    = YES
+
+# If you use Microsoft's C++/CLI language, you should set this option to YES to
+# enable parsing support.
+# The default value is: NO.
+
+CPP_CLI_SUPPORT        = NO
+
+# Set the SIP_SUPPORT tag to YES if your project consists of sip (see:
+# http://www.riverbankcomputing.co.uk/software/sip/intro) sources only. Doxygen
+# will parse them like normal C++ but will assume all classes use public instead
+# of private inheritance when no explicit protection keyword is present.
+# The default value is: NO.
+
+SIP_SUPPORT            = NO
+
+# For Microsoft's IDL there are propget and propput attributes to indicate
+# getter and setter methods for a property. Setting this option to YES will make
+# doxygen to replace the get and set methods by a property in the documentation.
+# This will only work if the methods are indeed getting or setting a simple
+# type. If this is not the case, or you want to show the methods anyway, you
+# should set this option to NO.
+# The default value is: YES.
+
+IDL_PROPERTY_SUPPORT   = YES
+
+# If member grouping is used in the documentation and the DISTRIBUTE_GROUP_DOC
+# tag is set to YES then doxygen will reuse the documentation of the first
+# member in the group (if any) for the other members of the group. By default
+# all members of a group must be documented explicitly.
+# The default value is: NO.
+
+DISTRIBUTE_GROUP_DOC   = YES
+
+# If one adds a struct or class to a group and this option is enabled, then also
+# any nested class or struct is added to the same group. By default this option
+# is disabled and one has to add nested compounds explicitly via \ingroup.
+# The default value is: NO.
+
+GROUP_NESTED_COMPOUNDS = NO
+
+# Set the SUBGROUPING tag to YES to allow class member groups of the same type
+# (for instance a group of public functions) to be put as a subgroup of that
+# type (e.g. under the Public Functions section). Set it to NO to prevent
+# subgrouping. Alternatively, this can be done per class using the
+# \nosubgrouping command.
+# The default value is: YES.
+
+SUBGROUPING            = YES
+
+# When the INLINE_GROUPED_CLASSES tag is set to YES, classes, structs and unions
+# are shown inside the group in which they are included (e.g. using \ingroup)
+# instead of on a separate page (for HTML and Man pages) or section (for LaTeX
+# and RTF).
+#
+# Note that this feature does not work in combination with
+# SEPARATE_MEMBER_PAGES.
+# The default value is: NO.
+
+INLINE_GROUPED_CLASSES = NO
+
+# When the INLINE_SIMPLE_STRUCTS tag is set to YES, structs, classes, and unions
+# with only public data fields or simple typedef fields will be shown inline in
+# the documentation of the scope in which they are defined (i.e. file,
+# namespace, or group documentation), provided this scope is documented. If set
+# to NO, structs, classes, and unions are shown on a separate page (for HTML and
+# Man pages) or section (for LaTeX and RTF).
+# The default value is: NO.
+
+INLINE_SIMPLE_STRUCTS  = NO
+
+# When TYPEDEF_HIDES_STRUCT tag is enabled, a typedef of a struct, union, or
+# enum is documented as struct, union, or enum with the name of the typedef. So
+# typedef struct TypeS {} TypeT, will appear in the documentation as a struct
+# with name TypeT. When disabled the typedef will appear as a member of a file,
+# namespace, or class. And the struct will be named TypeS. This can typically be
+# useful for C code in case the coding convention dictates that all compound
+# types are typedef'ed and only the typedef is referenced, never the tag name.
+# The default value is: NO.
+
+TYPEDEF_HIDES_STRUCT   = YES
+
+# The size of the symbol lookup cache can be set using LOOKUP_CACHE_SIZE. This
+# cache is used to resolve symbols given their name and scope. Since this can be
+# an expensive process and often the same symbol appears multiple times in the
+# code, doxygen keeps a cache of pre-resolved symbols. If the cache is too small
+# doxygen will become slower. If the cache is too large, memory is wasted. The
+# cache size is given by this formula: 2^(16+LOOKUP_CACHE_SIZE). The valid range
+# is 0..9, the default is 0, corresponding to a cache size of 2^16=65536
+# symbols. At the end of a run doxygen will report the cache usage and suggest
+# the optimal cache size from a speed point of view.
+# Minimum value: 0, maximum value: 9, default value: 0.
+
+LOOKUP_CACHE_SIZE      = 0
+
+#---------------------------------------------------------------------------
+# Build related configuration options
+#---------------------------------------------------------------------------
+
+# If the EXTRACT_ALL tag is set to YES, doxygen will assume all entities in
+# documentation are documented, even if no documentation was available. Private
+# class members and static file members will be hidden unless the
+# EXTRACT_PRIVATE respectively EXTRACT_STATIC tags are set to YES.
+# Note: This will also disable the warnings about undocumented members that are
+# normally produced when WARNINGS is set to YES.
+# The default value is: NO.
+
+EXTRACT_ALL            = YES
+
+# If the EXTRACT_PRIVATE tag is set to YES, all private members of a class will
+# be included in the documentation.
+# The default value is: NO.
+
+EXTRACT_PRIVATE        = NO
+
+# If the EXTRACT_PACKAGE tag is set to YES, all members with package or internal
+# scope will be included in the documentation.
+# The default value is: NO.
+
+EXTRACT_PACKAGE        = NO
+
+# If the EXTRACT_STATIC tag is set to YES, all static members of a file will be
+# included in the documentation.
+# The default value is: NO.
+
+EXTRACT_STATIC         = NO
+
+# If the EXTRACT_LOCAL_CLASSES tag is set to YES, classes (and structs) defined
+# locally in source files will be included in the documentation. If set to NO,
+# only classes defined in header files are included. Does not have any effect
+# for Java sources.
+# The default value is: YES.
+
+EXTRACT_LOCAL_CLASSES  = YES
+
+# This flag is only useful for Objective-C code. If set to YES, local methods,
+# which are defined in the implementation section but not in the interface are
+# included in the documentation. If set to NO, only methods in the interface are
+# included.
+# The default value is: NO.
+
+EXTRACT_LOCAL_METHODS  = NO
+
+# If this flag is set to YES, the members of anonymous namespaces will be
+# extracted and appear in the documentation as a namespace called
+# 'anonymous_namespace{file}', where file will be replaced with the base name of
+# the file that contains the anonymous namespace. By default anonymous namespace
+# are hidden.
+# The default value is: NO.
+
+EXTRACT_ANON_NSPACES   = NO
+
+# If the HIDE_UNDOC_MEMBERS tag is set to YES, doxygen will hide all
+# undocumented members inside documented classes or files. If set to NO these
+# members will be included in the various overviews, but no documentation
+# section is generated. This option has no effect if EXTRACT_ALL is enabled.
+# The default value is: NO.
+
+HIDE_UNDOC_MEMBERS     = NO
+
+# If the HIDE_UNDOC_CLASSES tag is set to YES, doxygen will hide all
+# undocumented classes that are normally visible in the class hierarchy. If set
+# to NO, these classes will be included in the various overviews. This option
+# has no effect if EXTRACT_ALL is enabled.
+# The default value is: NO.
+
+HIDE_UNDOC_CLASSES     = NO
+
+# If the HIDE_FRIEND_COMPOUNDS tag is set to YES, doxygen will hide all friend
+# (class|struct|union) declarations. If set to NO, these declarations will be
+# included in the documentation.
+# The default value is: NO.
+
+HIDE_FRIEND_COMPOUNDS  = NO
+
+# If the HIDE_IN_BODY_DOCS tag is set to YES, doxygen will hide any
+# documentation blocks found inside the body of a function. If set to NO, these
+# blocks will be appended to the function's detailed documentation block.
+# The default value is: NO.
+
+HIDE_IN_BODY_DOCS      = NO
+
+# The INTERNAL_DOCS tag determines if documentation that is typed after a
+# \internal command is included. If the tag is set to NO then the documentation
+# will be excluded. Set it to YES to include the internal documentation.
+# The default value is: NO.
+
+INTERNAL_DOCS          = NO
+
+# If the CASE_SENSE_NAMES tag is set to NO then doxygen will only generate file
+# names in lower-case letters. If set to YES, upper-case letters are also
+# allowed. This is useful if you have classes or files whose names only differ
+# in case and if your file system supports case sensitive file names. Windows
+# and Mac users are advised to set this option to NO.
+# The default value is: system dependent.
+
+CASE_SENSE_NAMES       = NO
+
+# If the HIDE_SCOPE_NAMES tag is set to NO then doxygen will show members with
+# their full class and namespace scopes in the documentation. If set to YES, the
+# scope will be hidden.
+# The default value is: NO.
+
+HIDE_SCOPE_NAMES       = NO
+
+# If the HIDE_COMPOUND_REFERENCE tag is set to NO (default) then doxygen will
+# append additional text to a page's title, such as Class Reference. If set to
+# YES the compound reference will be hidden.
+# The default value is: NO.
+
+HIDE_COMPOUND_REFERENCE= NO
+
+# If the SHOW_INCLUDE_FILES tag is set to YES then doxygen will put a list of
+# the files that are included by a file in the documentation of that file.
+# The default value is: YES.
+
+SHOW_INCLUDE_FILES     = YES
+
+# If the SHOW_GROUPED_MEMB_INC tag is set to YES then Doxygen will add for each
+# grouped member an include statement to the documentation, telling the reader
+# which file to include in order to use the member.
+# The default value is: NO.
+
+SHOW_GROUPED_MEMB_INC  = NO
+
+# If the FORCE_LOCAL_INCLUDES tag is set to YES then doxygen will list include
+# files with double quotes in the documentation rather than with sharp brackets.
+# The default value is: NO.
+
+FORCE_LOCAL_INCLUDES   = NO
+
+# If the INLINE_INFO tag is set to YES then a tag [inline] is inserted in the
+# documentation for inline members.
+# The default value is: YES.
+
+INLINE_INFO            = YES
+
+# If the SORT_MEMBER_DOCS tag is set to YES then doxygen will sort the
+# (detailed) documentation of file and class members alphabetically by member
+# name. If set to NO, the members will appear in declaration order.
+# The default value is: YES.
+
+SORT_MEMBER_DOCS       = YES
+
+# If the SORT_BRIEF_DOCS tag is set to YES then doxygen will sort the brief
+# descriptions of file, namespace and class members alphabetically by member
+# name. If set to NO, the members will appear in declaration order. Note that
+# this will also influence the order of the classes in the class list.
+# The default value is: NO.
+
+SORT_BRIEF_DOCS        = NO
+
+# If the SORT_MEMBERS_CTORS_1ST tag is set to YES then doxygen will sort the
+# (brief and detailed) documentation of class members so that constructors and
+# destructors are listed first. If set to NO the constructors will appear in the
+# respective orders defined by SORT_BRIEF_DOCS and SORT_MEMBER_DOCS.
+# Note: If SORT_BRIEF_DOCS is set to NO this option is ignored for sorting brief
+# member documentation.
+# Note: If SORT_MEMBER_DOCS is set to NO this option is ignored for sorting
+# detailed member documentation.
+# The default value is: NO.
+
+SORT_MEMBERS_CTORS_1ST = NO
+
+# If the SORT_GROUP_NAMES tag is set to YES then doxygen will sort the hierarchy
+# of group names into alphabetical order. If set to NO the group names will
+# appear in their defined order.
+# The default value is: NO.
+
+SORT_GROUP_NAMES       = NO
+
+# If the SORT_BY_SCOPE_NAME tag is set to YES, the class list will be sorted by
+# fully-qualified names, including namespaces. If set to NO, the class list will
+# be sorted only by class name, not including the namespace part.
+# Note: This option is not very useful if HIDE_SCOPE_NAMES is set to YES.
+# Note: This option applies only to the class list, not to the alphabetical
+# list.
+# The default value is: NO.
+
+SORT_BY_SCOPE_NAME     = NO
+
+# If the STRICT_PROTO_MATCHING option is enabled and doxygen fails to do proper
+# type resolution of all parameters of a function it will reject a match between
+# the prototype and the implementation of a member function even if there is
+# only one candidate or it is obvious which candidate to choose by doing a
+# simple string match. By disabling STRICT_PROTO_MATCHING doxygen will still
+# accept a match between prototype and implementation in such cases.
+# The default value is: NO.
+
+STRICT_PROTO_MATCHING  = NO
+
+# The GENERATE_TODOLIST tag can be used to enable (YES) or disable (NO) the todo
+# list. This list is created by putting \todo commands in the documentation.
+# The default value is: YES.
+
+GENERATE_TODOLIST      = YES
+
+# The GENERATE_TESTLIST tag can be used to enable (YES) or disable (NO) the test
+# list. This list is created by putting \test commands in the documentation.
+# The default value is: YES.
+
+GENERATE_TESTLIST      = YES
+
+# The GENERATE_BUGLIST tag can be used to enable (YES) or disable (NO) the bug
+# list. This list is created by putting \bug commands in the documentation.
+# The default value is: YES.
+
+GENERATE_BUGLIST       = YES
+
+# The GENERATE_DEPRECATEDLIST tag can be used to enable (YES) or disable (NO)
+# the deprecated list. This list is created by putting \deprecated commands in
+# the documentation.
+# The default value is: YES.
+
+GENERATE_DEPRECATEDLIST= YES
+
+# The ENABLED_SECTIONS tag can be used to enable conditional documentation
+# sections, marked by \if <section_label> ... \endif and \cond <section_label>
+# ... \endcond blocks.
+
+ENABLED_SECTIONS       =
+
+# The MAX_INITIALIZER_LINES tag determines the maximum number of lines that the
+# initial value of a variable or macro / define can have for it to appear in the
+# documentation. If the initializer consists of more lines than specified here
+# it will be hidden. Use a value of 0 to hide initializers completely. The
+# appearance of the value of individual variables and macros / defines can be
+# controlled using \showinitializer or \hideinitializer command in the
+# documentation regardless of this setting.
+# Minimum value: 0, maximum value: 10000, default value: 30.
+
+MAX_INITIALIZER_LINES  = 30
+
+# Set the SHOW_USED_FILES tag to NO to disable the list of files generated at
+# the bottom of the documentation of classes and structs. If set to YES, the
+# list will mention the files that were used to generate the documentation.
+# The default value is: YES.
+
+SHOW_USED_FILES        = YES
+
+# Set the SHOW_FILES tag to NO to disable the generation of the Files page. This
+# will remove the Files entry from the Quick Index and from the Folder Tree View
+# (if specified).
+# The default value is: YES.
+
+SHOW_FILES             = YES
+
+# Set the SHOW_NAMESPACES tag to NO to disable the generation of the Namespaces
+# page. This will remove the Namespaces entry from the Quick Index and from the
+# Folder Tree View (if specified).
+# The default value is: YES.
+
+SHOW_NAMESPACES        = YES
+
+# The FILE_VERSION_FILTER tag can be used to specify a program or script that
+# doxygen should invoke to get the current version for each file (typically from
+# the version control system). Doxygen will invoke the program by executing (via
+# popen()) the command command input-file, where command is the value of the
+# FILE_VERSION_FILTER tag, and input-file is the name of an input file provided
+# by doxygen. Whatever the program writes to standard output is used as the file
+# version. For an example see the documentation.
+
+FILE_VERSION_FILTER    =
+
+# The LAYOUT_FILE tag can be used to specify a layout file which will be parsed
+# by doxygen. The layout file controls the global structure of the generated
+# output files in an output format independent way. To create the layout file
+# that represents doxygen's defaults, run doxygen with the -l option. You can
+# optionally specify a file name after the option, if omitted DoxygenLayout.xml
+# will be used as the name of the layout file.
+#
+# Note that if you run doxygen from a directory containing a file called
+# DoxygenLayout.xml, doxygen will parse it automatically even if the LAYOUT_FILE
+# tag is left empty.
+
+LAYOUT_FILE            =
+
+# The CITE_BIB_FILES tag can be used to specify one or more bib files containing
+# the reference definitions. This must be a list of .bib files. The .bib
+# extension is automatically appended if omitted. This requires the bibtex tool
+# to be installed. See also http://en.wikipedia.org/wiki/BibTeX for more info.
+# For LaTeX the style of the bibliography can be controlled using
+# LATEX_BIB_STYLE. To use this feature you need bibtex and perl available in the
+# search path. See also \cite for info how to create references.
+
+CITE_BIB_FILES         =
+
+#---------------------------------------------------------------------------
+# Configuration options related to warning and progress messages
+#---------------------------------------------------------------------------
+
+# The QUIET tag can be used to turn on/off the messages that are generated to
+# standard output by doxygen. If QUIET is set to YES this implies that the
+# messages are off.
+# The default value is: NO.
+
+QUIET                  = NO
+
+# The WARNINGS tag can be used to turn on/off the warning messages that are
+# generated to standard error (stderr) by doxygen. If WARNINGS is set to YES
+# this implies that the warnings are on.
+#
+# Tip: Turn warnings on while writing the documentation.
+# The default value is: YES.
+
+WARNINGS               = YES
+
+# If the WARN_IF_UNDOCUMENTED tag is set to YES then doxygen will generate
+# warnings for undocumented members. If EXTRACT_ALL is set to YES then this flag
+# will automatically be disabled.
+# The default value is: YES.
+
+WARN_IF_UNDOCUMENTED   = YES
+
+# If the WARN_IF_DOC_ERROR tag is set to YES, doxygen will generate warnings for
+# potential errors in the documentation, such as not documenting some parameters
+# in a documented function, or documenting parameters that don't exist or using
+# markup commands wrongly.
+# The default value is: YES.
+
+WARN_IF_DOC_ERROR      = YES
+
+# This WARN_NO_PARAMDOC option can be enabled to get warnings for functions that
+# are documented, but have no documentation for their parameters or return
+# value. If set to NO, doxygen will only warn about wrong or incomplete
+# parameter documentation, but not about the absence of documentation.
+# The default value is: NO.
+
+WARN_NO_PARAMDOC       = NO
+
+# The WARN_FORMAT tag determines the format of the warning messages that doxygen
+# can produce. The string should contain the $file, $line, and $text tags, which
+# will be replaced by the file and line number from which the warning originated
+# and the warning text. Optionally the format may contain $version, which will
+# be replaced by the version of the file (if it could be obtained via
+# FILE_VERSION_FILTER)
+# The default value is: $file:$line: $text.
+
+WARN_FORMAT            = "$file:$line: $text"
+
+# The WARN_LOGFILE tag can be used to specify a file to which warning and error
+# messages should be written. If left blank the output is written to standard
+# error (stderr).
+
+WARN_LOGFILE           =
+
+#---------------------------------------------------------------------------
+# Configuration options related to the input files
+#---------------------------------------------------------------------------
+
+# The INPUT tag is used to specify the files and/or directories that contain
+# documented source files. You may enter file names like myfile.cpp or
+# directories like /usr/src/myproject. Separate the files or directories with
+# spaces. See also FILE_PATTERNS and EXTENSION_MAPPING
+# Note: If this tag is empty the current directory is searched.
+
+INPUT                  = ../../include/ck/tensor_operation/gpu/grid \
+                         ../../include/ck/tensor_operation/gpu/block \
+                         ../../include/ck/tensor_operation/gpu/thread \
+                         ../../library/include/ck/library/utility
+
+# This tag can be used to specify the character encoding of the source files
+# that doxygen parses. Internally doxygen uses the UTF-8 encoding. Doxygen uses
+# libiconv (or the iconv built into libc) for the transcoding. See the libiconv
+# documentation (see: http://www.gnu.org/software/libiconv) for the list of
+# possible encodings.
+# The default value is: UTF-8.
+
+INPUT_ENCODING         = UTF-8
+
+# If the value of the INPUT tag contains directories, you can use the
+# FILE_PATTERNS tag to specify one or more wildcard patterns (like *.cpp and
+# *.h) to filter out the source-files in the directories.
+#
+# Note that for custom extensions or not directly supported extensions you also
+# need to set EXTENSION_MAPPING for the extension otherwise the files are not
+# read by doxygen.
+#
+# If left blank the following patterns are tested:*.c, *.cc, *.cxx, *.cpp,
+# *.c++, *.java, *.ii, *.ixx, *.ipp, *.i++, *.inl, *.idl, *.ddl, *.odl, *.h,
+# *.hh, *.hxx, *.hpp, *.h++, *.cs, *.d, *.php, *.php4, *.php5, *.phtml, *.inc,
+# *.m, *.markdown, *.md, *.mm, *.dox, *.py, *.f90, *.f, *.for, *.tcl, *.vhd,
+# *.vhdl, *.ucf, *.qsf, *.as and *.js.
+
+FILE_PATTERNS          = *.c \
+                         *.cc \
+                         *.cxx \
+                         *.cpp \
+                         *.c++ \
+                         *.java \
+                         *.ii \
+                         *.ixx \
+                         *.ipp \
+                         *.i++ \
+                         *.inl \
+                         *.idl \
+                         *.ddl \
+                         *.odl \
+                         *.h \
+                         *.hh \
+                         *.hxx \
+                         *.hpp \
+                         *.h++ \
+                         *.cs \
+                         *.d \
+                         *.php \
+                         *.php4 \
+                         *.php5 \
+                         *.phtml \
+                         *.inc \
+                         *.m \
+                         *.markdown \
+                         *.md \
+                         *.mm \
+                         *.dox \
+                         *.py \
+                         *.tcl \
+                         *.vhd \
+                         *.vhdl \
+                         *.ucf \
+                         *.qsf \
+                         *.as \
+                         *.js
+
+# The RECURSIVE tag can be used to specify whether or not subdirectories should
+# be searched for input files as well.
+# The default value is: NO.
+
+RECURSIVE              = YES
+
+# The EXCLUDE tag can be used to specify files and/or directories that should be
+# excluded from the INPUT source files. This way you can easily exclude a
+# subdirectory from a directory tree whose root is specified with the INPUT tag.
+#
+# Note that relative paths are relative to the directory from which doxygen is
+# run.
+
+EXCLUDE                =
+
+# The EXCLUDE_SYMLINKS tag can be used to select whether or not files or
+# directories that are symbolic links (a Unix file system feature) are excluded
+# from the input.
+# The default value is: NO.
+
+EXCLUDE_SYMLINKS       = NO
+
+# If the value of the INPUT tag contains directories, you can use the
+# EXCLUDE_PATTERNS tag to specify one or more wildcard patterns to exclude
+# certain files from those directories.
+#
+# Note that the wildcards are matched against the file with absolute path, so to
+# exclude all test directories for example use the pattern */test/*
+
+EXCLUDE_PATTERNS       =
+
+# The EXCLUDE_SYMBOLS tag can be used to specify one or more symbol names
+# (namespaces, classes, functions, etc.) that should be excluded from the
+# output. The symbol name can be a fully qualified name, a word, or if the
+# wildcard * is used, a substring. Examples: ANamespace, AClass,
+# AClass::ANamespace, ANamespace::*Test
+#
+# Note that the wildcards are matched against the file with absolute path, so to
+# exclude all test directories use the pattern */test/*
+
+EXCLUDE_SYMBOLS        =
+
+# The EXAMPLE_PATH tag can be used to specify one or more files or directories
+# that contain example code fragments that are included (see the \include
+# command).
+
+EXAMPLE_PATH           =
+
+# If the value of the EXAMPLE_PATH tag contains directories, you can use the
+# EXAMPLE_PATTERNS tag to specify one or more wildcard pattern (like *.cpp and
+# *.h) to filter out the source-files in the directories. If left blank all
+# files are included.
+
+EXAMPLE_PATTERNS       = *
+
+# If the EXAMPLE_RECURSIVE tag is set to YES then subdirectories will be
+# searched for input files to be used with the \include or \dontinclude commands
+# irrespective of the value of the RECURSIVE tag.
+# The default value is: NO.
+
+EXAMPLE_RECURSIVE      = NO
+
+# The IMAGE_PATH tag can be used to specify one or more files or directories
+# that contain images that are to be included in the documentation (see the
+# \image command).
+
+IMAGE_PATH             =
+
+# The INPUT_FILTER tag can be used to specify a program that doxygen should
+# invoke to filter for each input file. Doxygen will invoke the filter program
+# by executing (via popen()) the command:
+#
+# <filter> <input-file>
+#
+# where <filter> is the value of the INPUT_FILTER tag, and <input-file> is the
+# name of an input file. Doxygen will then use the output that the filter
+# program writes to standard output. If FILTER_PATTERNS is specified, this tag
+# will be ignored.
+#
+# Note that the filter must not add or remove lines; it is applied before the
+# code is scanned, but not when the output code is generated. If lines are added
+# or removed, the anchors will not be placed correctly.
+
+INPUT_FILTER           =
+
+# The FILTER_PATTERNS tag can be used to specify filters on a per file pattern
+# basis. Doxygen will compare the file name with each pattern and apply the
+# filter if there is a match. The filters are a list of the form: pattern=filter
+# (like *.cpp=my_cpp_filter). See INPUT_FILTER for further information on how
+# filters are used. If the FILTER_PATTERNS tag is empty or if none of the
+# patterns match the file name, INPUT_FILTER is applied.
+
+FILTER_PATTERNS        =
+
+# If the FILTER_SOURCE_FILES tag is set to YES, the input filter (if set using
+# INPUT_FILTER) will also be used to filter the input files that are used for
+# producing the source files to browse (i.e. when SOURCE_BROWSER is set to YES).
+# The default value is: NO.
+
+FILTER_SOURCE_FILES    = NO
+
+# The FILTER_SOURCE_PATTERNS tag can be used to specify source filters per file
+# pattern. A pattern will override the setting for FILTER_PATTERN (if any) and
+# it is also possible to disable source filtering for a specific pattern using
+# *.ext= (so without naming a filter).
+# This tag requires that the tag FILTER_SOURCE_FILES is set to YES.
+
+FILTER_SOURCE_PATTERNS =
+
+# If the USE_MDFILE_AS_MAINPAGE tag refers to the name of a markdown file that
+# is part of the input, its contents will be placed on the main page
+# (index.html). This can be useful if you have a project on for instance GitHub
+# and want to reuse the introduction page also for the doxygen output.
+
+USE_MDFILE_AS_MAINPAGE = ../README.md
+
+#---------------------------------------------------------------------------
+# Configuration options related to source browsing
+#---------------------------------------------------------------------------
+
+# If the SOURCE_BROWSER tag is set to YES then a list of source files will be
+# generated. Documented entities will be cross-referenced with these sources.
+#
+# Note: To get rid of all source code in the generated output, make sure that
+# also VERBATIM_HEADERS is set to NO.
+# The default value is: NO.
+
+SOURCE_BROWSER         = NO
+
+# Setting the INLINE_SOURCES tag to YES will include the body of functions,
+# classes and enums directly into the documentation.
+# The default value is: NO.
+
+INLINE_SOURCES         = NO
+
+# Setting the STRIP_CODE_COMMENTS tag to YES will instruct doxygen to hide any
+# special comment blocks from generated source code fragments. Normal C, C++ and
+# Fortran comments will always remain visible.
+# The default value is: YES.
+
+STRIP_CODE_COMMENTS    = YES
+
+# If the REFERENCED_BY_RELATION tag is set to YES then for each documented
+# function all documented functions referencing it will be listed.
+# The default value is: NO.
+
+REFERENCED_BY_RELATION = NO
+
+# If the REFERENCES_RELATION tag is set to YES then for each documented function
+# all documented entities called/used by that function will be listed.
+# The default value is: NO.
+
+REFERENCES_RELATION    = NO
+
+# If the REFERENCES_LINK_SOURCE tag is set to YES and SOURCE_BROWSER tag is set
+# to YES then the hyperlinks from functions in REFERENCES_RELATION and
+# REFERENCED_BY_RELATION lists will link to the source code. Otherwise they will
+# link to the documentation.
+# The default value is: YES.
+
+REFERENCES_LINK_SOURCE = YES
+
+# If SOURCE_TOOLTIPS is enabled (the default) then hovering a hyperlink in the
+# source code will show a tooltip with additional information such as prototype,
+# brief description and links to the definition and documentation. Since this
+# will make the HTML file larger and loading of large files a bit slower, you
+# can opt to disable this feature.
+# The default value is: YES.
+# This tag requires that the tag SOURCE_BROWSER is set to YES.
+
+SOURCE_TOOLTIPS        = YES
+
+# If the USE_HTAGS tag is set to YES then the references to source code will
+# point to the HTML generated by the htags(1) tool instead of doxygen built-in
+# source browser. The htags tool is part of GNU's global source tagging system
+# (see http://www.gnu.org/software/global/global.html). You will need version
+# 4.8.6 or higher.
+#
+# To use it do the following:
+# - Install the latest version of global
+# - Enable SOURCE_BROWSER and USE_HTAGS in the config file
+# - Make sure the INPUT points to the root of the source tree
+# - Run doxygen as normal
+#
+# Doxygen will invoke htags (and that will in turn invoke gtags), so these
+# tools must be available from the command line (i.e. in the search path).
+#
+# The result: instead of the source browser generated by doxygen, the links to
+# source code will now point to the output of htags.
+# The default value is: NO.
+# This tag requires that the tag SOURCE_BROWSER is set to YES.
+
+USE_HTAGS              = NO
+
+# If the VERBATIM_HEADERS tag is set the YES then doxygen will generate a
+# verbatim copy of the header file for each class for which an include is
+# specified. Set to NO to disable this.
+# See also: Section \class.
+# The default value is: YES.
+
+VERBATIM_HEADERS       = YES
+
+# If the CLANG_ASSISTED_PARSING tag is set to YES then doxygen will use the
+# clang parser (see: http://clang.llvm.org/) for more accurate parsing at the
+# cost of reduced performance. This can be particularly helpful with template
+# rich C++ code for which doxygen's built-in parser lacks the necessary type
+# information.
+# Note: The availability of this option depends on whether or not doxygen was
+# compiled with the --with-libclang option.
+# The default value is: NO.
+
+CLANG_ASSISTED_PARSING = NO
+
+# If clang assisted parsing is enabled you can provide the compiler with command
+# line options that you would normally use when invoking the compiler. Note that
+# the include paths will already be set by doxygen for the files and directories
+# specified with INPUT and INCLUDE_PATH.
+# This tag requires that the tag CLANG_ASSISTED_PARSING is set to YES.
+
+CLANG_OPTIONS          =
+
+#---------------------------------------------------------------------------
+# Configuration options related to the alphabetical class index
+#---------------------------------------------------------------------------
+
+# If the ALPHABETICAL_INDEX tag is set to YES, an alphabetical index of all
+# compounds will be generated. Enable this if the project contains a lot of
+# classes, structs, unions or interfaces.
+# The default value is: YES.
+
+ALPHABETICAL_INDEX     = YES
+
+# The COLS_IN_ALPHA_INDEX tag can be used to specify the number of columns in
+# which the alphabetical index list will be split.
+# Minimum value: 1, maximum value: 20, default value: 5.
+# This tag requires that the tag ALPHABETICAL_INDEX is set to YES.
+
+COLS_IN_ALPHA_INDEX    = 5
+
+# In case all classes in a project start with a common prefix, all classes will
+# be put under the same header in the alphabetical index. The IGNORE_PREFIX tag
+# can be used to specify a prefix (or a list of prefixes) that should be ignored
+# while generating the index headers.
+# This tag requires that the tag ALPHABETICAL_INDEX is set to YES.
+
+IGNORE_PREFIX          =
+
+#---------------------------------------------------------------------------
+# Configuration options related to the HTML output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_HTML tag is set to YES, doxygen will generate HTML output
+# The default value is: YES.
+
+GENERATE_HTML          = YES
+
+# The HTML_OUTPUT tag is used to specify where the HTML docs will be put. If a
+# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
+# it.
+# The default directory is: html.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_OUTPUT            = html
+
+# The HTML_FILE_EXTENSION tag can be used to specify the file extension for each
+# generated HTML page (for example: .htm, .php, .asp).
+# The default value is: .html.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_FILE_EXTENSION    = .html
+
+# The HTML_HEADER tag can be used to specify a user-defined HTML header file for
+# each generated HTML page. If the tag is left blank doxygen will generate a
+# standard header.
+#
+# To get valid HTML the header file that includes any scripts and style sheets
+# that doxygen needs, which is dependent on the configuration options used (e.g.
+# the setting GENERATE_TREEVIEW). It is highly recommended to start with a
+# default header using
+# doxygen -w html new_header.html new_footer.html new_stylesheet.css
+# YourConfigFile
+# and then modify the file new_header.html. See also section "Doxygen usage"
+# for information on how to generate the default header that doxygen normally
+# uses.
+# Note: The header is subject to change so you typically have to regenerate the
+# default header when upgrading to a newer version of doxygen. For a description
+# of the possible markers and block names see the documentation.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_HEADER            =
+
+# The HTML_FOOTER tag can be used to specify a user-defined HTML footer for each
+# generated HTML page. If the tag is left blank doxygen will generate a standard
+# footer. See HTML_HEADER for more information on how to generate a default
+# footer and what special commands can be used inside the footer. See also
+# section "Doxygen usage" for information on how to generate the default footer
+# that doxygen normally uses.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_FOOTER            =
+
+# The HTML_STYLESHEET tag can be used to specify a user-defined cascading style
+# sheet that is used by each HTML page. It can be used to fine-tune the look of
+# the HTML output. If left blank doxygen will generate a default style sheet.
+# See also section "Doxygen usage" for information on how to generate the style
+# sheet that doxygen normally uses.
+# Note: It is recommended to use HTML_EXTRA_STYLESHEET instead of this tag, as
+# it is more robust and this tag (HTML_STYLESHEET) will in the future become
+# obsolete.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_STYLESHEET        =
+
+# The HTML_EXTRA_STYLESHEET tag can be used to specify additional user-defined
+# cascading style sheets that are included after the standard style sheets
+# created by doxygen. Using this option one can overrule certain style aspects.
+# This is preferred over using HTML_STYLESHEET since it does not replace the
+# standard style sheet and is therefore more robust against future updates.
+# Doxygen will copy the style sheet files to the output directory.
+# Note: The order of the extra style sheet files is of importance (e.g. the last
+# style sheet in the list overrules the setting of the previous ones in the
+# list). For an example see the documentation.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_EXTRA_STYLESHEET  =
+
+# The HTML_EXTRA_FILES tag can be used to specify one or more extra images or
+# other source files which should be copied to the HTML output directory. Note
+# that these files will be copied to the base HTML output directory. Use the
+# $relpath^ marker in the HTML_HEADER and/or HTML_FOOTER files to load these
+# files. In the HTML_STYLESHEET file, use the file name only. Also note that the
+# files will be copied as-is; there are no commands or markers available.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_EXTRA_FILES       =
+
+# The HTML_COLORSTYLE_HUE tag controls the color of the HTML output. Doxygen
+# will adjust the colors in the style sheet and background images according to
+# this color. Hue is specified as an angle on a colorwheel, see
+# http://en.wikipedia.org/wiki/Hue for more information. For instance the value
+# 0 represents red, 60 is yellow, 120 is green, 180 is cyan, 240 is blue, 300
+# purple, and 360 is red again.
+# Minimum value: 0, maximum value: 359, default value: 220.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_COLORSTYLE_HUE    = 220
+
+# The HTML_COLORSTYLE_SAT tag controls the purity (or saturation) of the colors
+# in the HTML output. For a value of 0 the output will use grayscales only. A
+# value of 255 will produce the most vivid colors.
+# Minimum value: 0, maximum value: 255, default value: 100.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_COLORSTYLE_SAT    = 100
+
+# The HTML_COLORSTYLE_GAMMA tag controls the gamma correction applied to the
+# luminance component of the colors in the HTML output. Values below 100
+# gradually make the output lighter, whereas values above 100 make the output
+# darker. The value divided by 100 is the actual gamma applied, so 80 represents
+# a gamma of 0.8, The value 220 represents a gamma of 2.2, and 100 does not
+# change the gamma.
+# Minimum value: 40, maximum value: 240, default value: 80.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_COLORSTYLE_GAMMA  = 80
+
+# If the HTML_TIMESTAMP tag is set to YES then the footer of each generated HTML
+# page will contain the date and time when the page was generated. Setting this
+# to YES can help to show when doxygen was last run and thus if the
+# documentation is up to date.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_TIMESTAMP         = NO
+
+# If the HTML_DYNAMIC_SECTIONS tag is set to YES then the generated HTML
+# documentation will contain sections that can be hidden and shown after the
+# page has loaded.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_DYNAMIC_SECTIONS  = NO
+
+# With HTML_INDEX_NUM_ENTRIES one can control the preferred number of entries
+# shown in the various tree structured indices initially; the user can expand
+# and collapse entries dynamically later on. Doxygen will expand the tree to
+# such a level that at most the specified number of entries are visible (unless
+# a fully collapsed tree already exceeds this amount). So setting the number of
+# entries 1 will produce a full collapsed tree by default. 0 is a special value
+# representing an infinite number of entries and will result in a full expanded
+# tree by default.
+# Minimum value: 0, maximum value: 9999, default value: 100.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+HTML_INDEX_NUM_ENTRIES = 100
+
+# If the GENERATE_DOCSET tag is set to YES, additional index files will be
+# generated that can be used as input for Apple's Xcode 3 integrated development
+# environment (see: http://developer.apple.com/tools/xcode/), introduced with
+# OSX 10.5 (Leopard). To create a documentation set, doxygen will generate a
+# Makefile in the HTML output directory. Running make will produce the docset in
+# that directory and running make install will install the docset in
+# ~/Library/Developer/Shared/Documentation/DocSets so that Xcode will find it at
+# startup. See http://developer.apple.com/tools/creatingdocsetswithdoxygen.html
+# for more information.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+GENERATE_DOCSET        = NO
+
+# This tag determines the name of the docset feed. A documentation feed provides
+# an umbrella under which multiple documentation sets from a single provider
+# (such as a company or product suite) can be grouped.
+# The default value is: Doxygen generated docs.
+# This tag requires that the tag GENERATE_DOCSET is set to YES.
+
+DOCSET_FEEDNAME        = "Doxygen generated docs"
+
+# This tag specifies a string that should uniquely identify the documentation
+# set bundle. This should be a reverse domain-name style string, e.g.
+# com.mycompany.MyDocSet. Doxygen will append .docset to the name.
+# The default value is: org.doxygen.Project.
+# This tag requires that the tag GENERATE_DOCSET is set to YES.
+
+DOCSET_BUNDLE_ID       = org.doxygen.Project
+
+# The DOCSET_PUBLISHER_ID tag specifies a string that should uniquely identify
+# the documentation publisher. This should be a reverse domain-name style
+# string, e.g. com.mycompany.MyDocSet.documentation.
+# The default value is: org.doxygen.Publisher.
+# This tag requires that the tag GENERATE_DOCSET is set to YES.
+
+DOCSET_PUBLISHER_ID    = org.doxygen.Publisher
+
+# The DOCSET_PUBLISHER_NAME tag identifies the documentation publisher.
+# The default value is: Publisher.
+# This tag requires that the tag GENERATE_DOCSET is set to YES.
+
+DOCSET_PUBLISHER_NAME  = Publisher
+
+# If the GENERATE_HTMLHELP tag is set to YES then doxygen generates three
+# additional HTML index files: index.hhp, index.hhc, and index.hhk. The
+# index.hhp is a project file that can be read by Microsoft's HTML Help Workshop
+# (see: http://www.microsoft.com/en-us/download/details.aspx?id=21138) on
+# Windows.
+#
+# The HTML Help Workshop contains a compiler that can convert all HTML output
+# generated by doxygen into a single compiled HTML file (.chm). Compiled HTML
+# files are now used as the Windows 98 help format, and will replace the old
+# Windows help format (.hlp) on all Windows platforms in the future. Compressed
+# HTML files also contain an index, a table of contents, and you can search for
+# words in the documentation. The HTML workshop also contains a viewer for
+# compressed HTML files.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+GENERATE_HTMLHELP      = NO
+
+# The CHM_FILE tag can be used to specify the file name of the resulting .chm
+# file. You can add a path in front of the file if the result should not be
+# written to the html output directory.
+# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
+
+CHM_FILE               =
+
+# The HHC_LOCATION tag can be used to specify the location (absolute path
+# including file name) of the HTML help compiler (hhc.exe). If non-empty,
+# doxygen will try to run the HTML help compiler on the generated index.hhp.
+# The file has to be specified with full path.
+# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
+
+HHC_LOCATION           =
+
+# The GENERATE_CHI flag controls if a separate .chi index file is generated
+# (YES) or that it should be included in the master .chm file (NO).
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
+
+GENERATE_CHI           = NO
+
+# The CHM_INDEX_ENCODING is used to encode HtmlHelp index (hhk), content (hhc)
+# and project file content.
+# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
+
+CHM_INDEX_ENCODING     =
+
+# The BINARY_TOC flag controls whether a binary table of contents is generated
+# (YES) or a normal table of contents (NO) in the .chm file. Furthermore it
+# enables the Previous and Next buttons.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
+
+BINARY_TOC             = NO
+
+# The TOC_EXPAND flag can be set to YES to add extra items for group members to
+# the table of contents of the HTML help documentation and to the tree view.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
+
+TOC_EXPAND             = NO
+
+# If the GENERATE_QHP tag is set to YES and both QHP_NAMESPACE and
+# QHP_VIRTUAL_FOLDER are set, an additional index file will be generated that
+# can be used as input for Qt's qhelpgenerator to generate a Qt Compressed Help
+# (.qch) of the generated HTML documentation.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+GENERATE_QHP           = NO
+
+# If the QHG_LOCATION tag is specified, the QCH_FILE tag can be used to specify
+# the file name of the resulting .qch file. The path specified is relative to
+# the HTML output folder.
+# This tag requires that the tag GENERATE_QHP is set to YES.
+
+QCH_FILE               =
+
+# The QHP_NAMESPACE tag specifies the namespace to use when generating Qt Help
+# Project output. For more information please see Qt Help Project / Namespace
+# (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#namespace).
+# The default value is: org.doxygen.Project.
+# This tag requires that the tag GENERATE_QHP is set to YES.
+
+QHP_NAMESPACE          = org.doxygen.Project
+
+# The QHP_VIRTUAL_FOLDER tag specifies the namespace to use when generating Qt
+# Help Project output. For more information please see Qt Help Project / Virtual
+# Folders (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#virtual-
+# folders).
+# The default value is: doc.
+# This tag requires that the tag GENERATE_QHP is set to YES.
+
+QHP_VIRTUAL_FOLDER     = doc
+
+# If the QHP_CUST_FILTER_NAME tag is set, it specifies the name of a custom
+# filter to add. For more information please see Qt Help Project / Custom
+# Filters (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#custom-
+# filters).
+# This tag requires that the tag GENERATE_QHP is set to YES.
+
+QHP_CUST_FILTER_NAME   =
+
+# The QHP_CUST_FILTER_ATTRS tag specifies the list of the attributes of the
+# custom filter to add. For more information please see Qt Help Project / Custom
+# Filters (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#custom-
+# filters).
+# This tag requires that the tag GENERATE_QHP is set to YES.
+
+QHP_CUST_FILTER_ATTRS  =
+
+# The QHP_SECT_FILTER_ATTRS tag specifies the list of the attributes this
+# project's filter section matches. Qt Help Project / Filter Attributes (see:
+# http://qt-project.org/doc/qt-4.8/qthelpproject.html#filter-attributes).
+# This tag requires that the tag GENERATE_QHP is set to YES.
+
+QHP_SECT_FILTER_ATTRS  =
+
+# The QHG_LOCATION tag can be used to specify the location of Qt's
+# qhelpgenerator. If non-empty doxygen will try to run qhelpgenerator on the
+# generated .qhp file.
+# This tag requires that the tag GENERATE_QHP is set to YES.
+
+QHG_LOCATION           =
+
+# If the GENERATE_ECLIPSEHELP tag is set to YES, additional index files will be
+# generated, together with the HTML files, they form an Eclipse help plugin. To
+# install this plugin and make it available under the help contents menu in
+# Eclipse, the contents of the directory containing the HTML and XML files needs
+# to be copied into the plugins directory of eclipse. The name of the directory
+# within the plugins directory should be the same as the ECLIPSE_DOC_ID value.
+# After copying Eclipse needs to be restarted before the help appears.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+GENERATE_ECLIPSEHELP   = NO
+
+# A unique identifier for the Eclipse help plugin. When installing the plugin
+# the directory name containing the HTML and XML files should also have this
+# name. Each documentation set should have its own identifier.
+# The default value is: org.doxygen.Project.
+# This tag requires that the tag GENERATE_ECLIPSEHELP is set to YES.
+
+ECLIPSE_DOC_ID         = org.doxygen.Project
+
+# If you want full control over the layout of the generated HTML pages it might
+# be necessary to disable the index and replace it with your own. The
+# DISABLE_INDEX tag can be used to turn on/off the condensed index (tabs) at top
+# of each HTML page. A value of NO enables the index and the value YES disables
+# it. Since the tabs in the index contain the same information as the navigation
+# tree, you can set this option to YES if you also set GENERATE_TREEVIEW to YES.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+DISABLE_INDEX          = NO
+
+# The GENERATE_TREEVIEW tag is used to specify whether a tree-like index
+# structure should be generated to display hierarchical information. If the tag
+# value is set to YES, a side panel will be generated containing a tree-like
+# index structure (just like the one that is generated for HTML Help). For this
+# to work a browser that supports JavaScript, DHTML, CSS and frames is required
+# (i.e. any modern browser). Windows users are probably better off using the
+# HTML help feature. Via custom style sheets (see HTML_EXTRA_STYLESHEET) one can
+# further fine-tune the look of the index. As an example, the default style
+# sheet generated by doxygen has an example that shows how to put an image at
+# the root of the tree instead of the PROJECT_NAME. Since the tree basically has
+# the same information as the tab index, you could consider setting
+# DISABLE_INDEX to YES when enabling this option.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+GENERATE_TREEVIEW      = NO
+
+# The ENUM_VALUES_PER_LINE tag can be used to set the number of enum values that
+# doxygen will group on one line in the generated HTML documentation.
+#
+# Note that a value of 0 will completely suppress the enum values from appearing
+# in the overview section.
+# Minimum value: 0, maximum value: 20, default value: 4.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+ENUM_VALUES_PER_LINE   = 1
+
+# If the treeview is enabled (see GENERATE_TREEVIEW) then this tag can be used
+# to set the initial width (in pixels) of the frame in which the tree is shown.
+# Minimum value: 0, maximum value: 1500, default value: 250.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+TREEVIEW_WIDTH         = 250
+
+# If the EXT_LINKS_IN_WINDOW option is set to YES, doxygen will open links to
+# external symbols imported via tag files in a separate window.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+EXT_LINKS_IN_WINDOW    = NO
+
+# Use this tag to change the font size of LaTeX formulas included as images in
+# the HTML documentation. When you change the font size after a successful
+# doxygen run you need to manually remove any form_*.png images from the HTML
+# output directory to force them to be regenerated.
+# Minimum value: 8, maximum value: 50, default value: 10.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+FORMULA_FONTSIZE       = 10
+
+# Use the FORMULA_TRANPARENT tag to determine whether or not the images
+# generated for formulas are transparent PNGs. Transparent PNGs are not
+# supported properly for IE 6.0, but are supported on all modern browsers.
+#
+# Note that when changing this option you need to delete any form_*.png files in
+# the HTML output directory before the changes have effect.
+# The default value is: YES.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+FORMULA_TRANSPARENT    = YES
+
+# Enable the USE_MATHJAX option to render LaTeX formulas using MathJax (see
+# http://www.mathjax.org) which uses client side Javascript for the rendering
+# instead of using pre-rendered bitmaps. Use this if you do not have LaTeX
+# installed or if you want to formulas look prettier in the HTML output. When
+# enabled you may also need to install MathJax separately and configure the path
+# to it using the MATHJAX_RELPATH option.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+USE_MATHJAX            = YES
+
+# When MathJax is enabled you can set the default output format to be used for
+# the MathJax output. See the MathJax site (see:
+# http://docs.mathjax.org/en/latest/output.html) for more details.
+# Possible values are: HTML-CSS (which is slower, but has the best
+# compatibility), NativeMML (i.e. MathML) and SVG.
+# The default value is: HTML-CSS.
+# This tag requires that the tag USE_MATHJAX is set to YES.
+
+MATHJAX_FORMAT         = HTML-CSS
+
+# When MathJax is enabled you need to specify the location relative to the HTML
+# output directory using the MATHJAX_RELPATH option. The destination directory
+# should contain the MathJax.js script. For instance, if the mathjax directory
+# is located at the same level as the HTML output directory, then
+# MATHJAX_RELPATH should be ../mathjax. The default value points to the MathJax
+# Content Delivery Network so you can quickly see the result without installing
+# MathJax. However, it is strongly recommended to install a local copy of
+# MathJax from http://www.mathjax.org before deployment.
+# The default value is: http://cdn.mathjax.org/mathjax/latest.
+# This tag requires that the tag USE_MATHJAX is set to YES.
+
+MATHJAX_RELPATH        = http://cdn.mathjax.org/mathjax/latest
+
+# The MATHJAX_EXTENSIONS tag can be used to specify one or more MathJax
+# extension names that should be enabled during MathJax rendering. For example
+# MATHJAX_EXTENSIONS = TeX/AMSmath TeX/AMSsymbols
+# This tag requires that the tag USE_MATHJAX is set to YES.
+
+MATHJAX_EXTENSIONS     =
+
+# The MATHJAX_CODEFILE tag can be used to specify a file with javascript pieces
+# of code that will be used on startup of the MathJax code. See the MathJax site
+# (see: http://docs.mathjax.org/en/latest/output.html) for more details. For an
+# example see the documentation.
+# This tag requires that the tag USE_MATHJAX is set to YES.
+
+MATHJAX_CODEFILE       =
+
+# When the SEARCHENGINE tag is enabled doxygen will generate a search box for
+# the HTML output. The underlying search engine uses javascript and DHTML and
+# should work on any modern browser. Note that when using HTML help
+# (GENERATE_HTMLHELP), Qt help (GENERATE_QHP), or docsets (GENERATE_DOCSET)
+# there is already a search function so this one should typically be disabled.
+# For large projects the javascript based search engine can be slow, then
+# enabling SERVER_BASED_SEARCH may provide a better solution. It is possible to
+# search using the keyboard; to jump to the search box use <access key> + S
+# (what the <access key> is depends on the OS and browser, but it is typically
+# <CTRL>, <ALT>/<option>, or both). Inside the search box use the <cursor down
+# key> to jump into the search results window, the results can be navigated
+# using the <cursor keys>. Press <Enter> to select an item or <escape> to cancel
+# the search. The filter options can be selected when the cursor is inside the
+# search box by pressing <Shift>+<cursor down>. Also here use the <cursor keys>
+# to select a filter and <Enter> or <escape> to activate or cancel the filter
+# option.
+# The default value is: YES.
+# This tag requires that the tag GENERATE_HTML is set to YES.
+
+SEARCHENGINE           = YES
+
+# When the SERVER_BASED_SEARCH tag is enabled the search engine will be
+# implemented using a web server instead of a web client using Javascript. There
+# are two flavors of web server based searching depending on the EXTERNAL_SEARCH
+# setting. When disabled, doxygen will generate a PHP script for searching and
+# an index file used by the script. When EXTERNAL_SEARCH is enabled the indexing
+# and searching needs to be provided by external tools. See the section
+# "External Indexing and Searching" for details.
+# The default value is: NO.
+# This tag requires that the tag SEARCHENGINE is set to YES.
+
+SERVER_BASED_SEARCH    = NO
+
+# When EXTERNAL_SEARCH tag is enabled doxygen will no longer generate the PHP
+# script for searching. Instead the search results are written to an XML file
+# which needs to be processed by an external indexer. Doxygen will invoke an
+# external search engine pointed to by the SEARCHENGINE_URL option to obtain the
+# search results.
+#
+# Doxygen ships with an example indexer (doxyindexer) and search engine
+# (doxysearch.cgi) which are based on the open source search engine library
+# Xapian (see: http://xapian.org/).
+#
+# See the section "External Indexing and Searching" for details.
+# The default value is: NO.
+# This tag requires that the tag SEARCHENGINE is set to YES.
+
+EXTERNAL_SEARCH        = NO
+
+# The SEARCHENGINE_URL should point to a search engine hosted by a web server
+# which will return the search results when EXTERNAL_SEARCH is enabled.
+#
+# Doxygen ships with an example indexer (doxyindexer) and search engine
+# (doxysearch.cgi) which are based on the open source search engine library
+# Xapian (see: http://xapian.org/). See the section "External Indexing and
+# Searching" for details.
+# This tag requires that the tag SEARCHENGINE is set to YES.
+
+SEARCHENGINE_URL       =
+
+# When SERVER_BASED_SEARCH and EXTERNAL_SEARCH are both enabled the unindexed
+# search data is written to a file for indexing by an external tool. With the
+# SEARCHDATA_FILE tag the name of this file can be specified.
+# The default file is: searchdata.xml.
+# This tag requires that the tag SEARCHENGINE is set to YES.
+
+SEARCHDATA_FILE        = searchdata.xml
+
+# When SERVER_BASED_SEARCH and EXTERNAL_SEARCH are both enabled the
+# EXTERNAL_SEARCH_ID tag can be used as an identifier for the project. This is
+# useful in combination with EXTRA_SEARCH_MAPPINGS to search through multiple
+# projects and redirect the results back to the right project.
+# This tag requires that the tag SEARCHENGINE is set to YES.
+
+EXTERNAL_SEARCH_ID     =
+
+# The EXTRA_SEARCH_MAPPINGS tag can be used to enable searching through doxygen
+# projects other than the one defined by this configuration file, but that are
+# all added to the same external search index. Each project needs to have a
+# unique id set via EXTERNAL_SEARCH_ID. The search mapping then maps the id of
+# to a relative location where the documentation can be found. The format is:
+# EXTRA_SEARCH_MAPPINGS = tagname1=loc1 tagname2=loc2 ...
+# This tag requires that the tag SEARCHENGINE is set to YES.
+
+EXTRA_SEARCH_MAPPINGS  =
+
+#---------------------------------------------------------------------------
+# Configuration options related to the LaTeX output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_LATEX tag is set to YES, doxygen will generate LaTeX output.
+# The default value is: YES.
+
+GENERATE_LATEX         = NO
+
+# The LATEX_OUTPUT tag is used to specify where the LaTeX docs will be put. If a
+# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
+# it.
+# The default directory is: latex.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_OUTPUT           = latex
+
+# The LATEX_CMD_NAME tag can be used to specify the LaTeX command name to be
+# invoked.
+#
+# Note that when enabling USE_PDFLATEX this option is only used for generating
+# bitmaps for formulas in the HTML output, but not in the Makefile that is
+# written to the output directory.
+# The default file is: latex.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_CMD_NAME         = latex
+
+# The MAKEINDEX_CMD_NAME tag can be used to specify the command name to generate
+# index for LaTeX.
+# The default file is: makeindex.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+MAKEINDEX_CMD_NAME     = makeindex
+
+# If the COMPACT_LATEX tag is set to YES, doxygen generates more compact LaTeX
+# documents. This may be useful for small projects and may help to save some
+# trees in general.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+COMPACT_LATEX          = NO
+
+# The PAPER_TYPE tag can be used to set the paper type that is used by the
+# printer.
+# Possible values are: a4 (210 x 297 mm), letter (8.5 x 11 inches), legal (8.5 x
+# 14 inches) and executive (7.25 x 10.5 inches).
+# The default value is: a4.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+PAPER_TYPE             = a4
+
+# The EXTRA_PACKAGES tag can be used to specify one or more LaTeX package names
+# that should be included in the LaTeX output. The package can be specified just
+# by its name or with the correct syntax as to be used with the LaTeX
+# \usepackage command. To get the times font for instance you can specify :
+# EXTRA_PACKAGES=times or EXTRA_PACKAGES={times}
+# To use the option intlimits with the amsmath package you can specify:
+# EXTRA_PACKAGES=[intlimits]{amsmath}
+# If left blank no extra packages will be included.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+EXTRA_PACKAGES         =
+
+# The LATEX_HEADER tag can be used to specify a personal LaTeX header for the
+# generated LaTeX document. The header should contain everything until the first
+# chapter. If it is left blank doxygen will generate a standard header. See
+# section "Doxygen usage" for information on how to let doxygen write the
+# default header to a separate file.
+#
+# Note: Only use a user-defined header if you know what you are doing! The
+# following commands have a special meaning inside the header: $title,
+# $datetime, $date, $doxygenversion, $projectname, $projectnumber,
+# $projectbrief, $projectlogo. Doxygen will replace $title with the empty
+# string, for the replacement values of the other commands the user is referred
+# to HTML_HEADER.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_HEADER           =
+
+# The LATEX_FOOTER tag can be used to specify a personal LaTeX footer for the
+# generated LaTeX document. The footer should contain everything after the last
+# chapter. If it is left blank doxygen will generate a standard footer. See
+# LATEX_HEADER for more information on how to generate a default footer and what
+# special commands can be used inside the footer.
+#
+# Note: Only use a user-defined footer if you know what you are doing!
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_FOOTER           =
+
+# The LATEX_EXTRA_STYLESHEET tag can be used to specify additional user-defined
+# LaTeX style sheets that are included after the standard style sheets created
+# by doxygen. Using this option one can overrule certain style aspects. Doxygen
+# will copy the style sheet files to the output directory.
+# Note: The order of the extra style sheet files is of importance (e.g. the last
+# style sheet in the list overrules the setting of the previous ones in the
+# list).
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_EXTRA_STYLESHEET =
+
+# The LATEX_EXTRA_FILES tag can be used to specify one or more extra images or
+# other source files which should be copied to the LATEX_OUTPUT output
+# directory. Note that the files will be copied as-is; there are no commands or
+# markers available.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_EXTRA_FILES      =
+
+# If the PDF_HYPERLINKS tag is set to YES, the LaTeX that is generated is
+# prepared for conversion to PDF (using ps2pdf or pdflatex). The PDF file will
+# contain links (just like the HTML output) instead of page references. This
+# makes the output suitable for online browsing using a PDF viewer.
+# The default value is: YES.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+PDF_HYPERLINKS         = YES
+
+# If the USE_PDFLATEX tag is set to YES, doxygen will use pdflatex to generate
+# the PDF file directly from the LaTeX files. Set this option to YES, to get a
+# higher quality PDF documentation.
+# The default value is: YES.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+USE_PDFLATEX           = YES
+
+# If the LATEX_BATCHMODE tag is set to YES, doxygen will add the \batchmode
+# command to the generated LaTeX files. This will instruct LaTeX to keep running
+# if errors occur, instead of asking the user for help. This option is also used
+# when generating formulas in HTML.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_BATCHMODE        = NO
+
+# If the LATEX_HIDE_INDICES tag is set to YES then doxygen will not include the
+# index chapters (such as File Index, Compound Index, etc.) in the output.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_HIDE_INDICES     = NO
+
+# If the LATEX_SOURCE_CODE tag is set to YES then doxygen will include source
+# code with syntax highlighting in the LaTeX output.
+#
+# Note that which sources are shown also depends on other settings such as
+# SOURCE_BROWSER.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_SOURCE_CODE      = NO
+
+# The LATEX_BIB_STYLE tag can be used to specify the style to use for the
+# bibliography, e.g. plainnat, or ieeetr. See
+# http://en.wikipedia.org/wiki/BibTeX and \cite for more info.
+# The default value is: plain.
+# This tag requires that the tag GENERATE_LATEX is set to YES.
+
+LATEX_BIB_STYLE        = plain
+
+#---------------------------------------------------------------------------
+# Configuration options related to the RTF output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_RTF tag is set to YES, doxygen will generate RTF output. The
+# RTF output is optimized for Word 97 and may not look too pretty with other RTF
+# readers/editors.
+# The default value is: NO.
+
+GENERATE_RTF           = NO
+
+# The RTF_OUTPUT tag is used to specify where the RTF docs will be put. If a
+# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
+# it.
+# The default directory is: rtf.
+# This tag requires that the tag GENERATE_RTF is set to YES.
+
+RTF_OUTPUT             = rtf
+
+# If the COMPACT_RTF tag is set to YES, doxygen generates more compact RTF
+# documents. This may be useful for small projects and may help to save some
+# trees in general.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_RTF is set to YES.
+
+COMPACT_RTF            = NO
+
+# If the RTF_HYPERLINKS tag is set to YES, the RTF that is generated will
+# contain hyperlink fields. The RTF file will contain links (just like the HTML
+# output) instead of page references. This makes the output suitable for online
+# browsing using Word or some other Word compatible readers that support those
+# fields.
+#
+# Note: WordPad (write) and others do not support links.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_RTF is set to YES.
+
+RTF_HYPERLINKS         = NO
+
+# Load stylesheet definitions from file. Syntax is similar to doxygen's config
+# file, i.e. a series of assignments. You only have to provide replacements,
+# missing definitions are set to their default value.
+#
+# See also section "Doxygen usage" for information on how to generate the
+# default style sheet that doxygen normally uses.
+# This tag requires that the tag GENERATE_RTF is set to YES.
+
+RTF_STYLESHEET_FILE    =
+
+# Set optional variables used in the generation of an RTF document. Syntax is
+# similar to doxygen's config file. A template extensions file can be generated
+# using doxygen -e rtf extensionFile.
+# This tag requires that the tag GENERATE_RTF is set to YES.
+
+RTF_EXTENSIONS_FILE    =
+
+# If the RTF_SOURCE_CODE tag is set to YES then doxygen will include source code
+# with syntax highlighting in the RTF output.
+#
+# Note that which sources are shown also depends on other settings such as
+# SOURCE_BROWSER.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_RTF is set to YES.
+
+RTF_SOURCE_CODE        = NO
+
+#---------------------------------------------------------------------------
+# Configuration options related to the man page output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_MAN tag is set to YES, doxygen will generate man pages for
+# classes and files.
+# The default value is: NO.
+
+GENERATE_MAN           = NO
+
+# The MAN_OUTPUT tag is used to specify where the man pages will be put. If a
+# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
+# it. A directory man3 will be created inside the directory specified by
+# MAN_OUTPUT.
+# The default directory is: man.
+# This tag requires that the tag GENERATE_MAN is set to YES.
+
+MAN_OUTPUT             = man
+
+# The MAN_EXTENSION tag determines the extension that is added to the generated
+# man pages. In case the manual section does not start with a number, the number
+# 3 is prepended. The dot (.) at the beginning of the MAN_EXTENSION tag is
+# optional.
+# The default value is: .3.
+# This tag requires that the tag GENERATE_MAN is set to YES.
+
+MAN_EXTENSION          = .3
+
+# The MAN_SUBDIR tag determines the name of the directory created within
+# MAN_OUTPUT in which the man pages are placed. If defaults to man followed by
+# MAN_EXTENSION with the initial . removed.
+# This tag requires that the tag GENERATE_MAN is set to YES.
+
+MAN_SUBDIR             =
+
+# If the MAN_LINKS tag is set to YES and doxygen generates man output, then it
+# will generate one additional man file for each entity documented in the real
+# man page(s). These additional files only source the real man page, but without
+# them the man command would be unable to find the correct page.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_MAN is set to YES.
+
+MAN_LINKS              = NO
+
+#---------------------------------------------------------------------------
+# Configuration options related to the XML output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_XML tag is set to YES, doxygen will generate an XML file that
+# captures the structure of the code including all documentation.
+# The default value is: NO.
+
+GENERATE_XML           = YES
+
+# The XML_OUTPUT tag is used to specify where the XML pages will be put. If a
+# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
+# it.
+# The default directory is: xml.
+# This tag requires that the tag GENERATE_XML is set to YES.
+
+XML_OUTPUT             = xml
+
+# If the XML_PROGRAMLISTING tag is set to YES, doxygen will dump the program
+# listings (including syntax highlighting and cross-referencing information) to
+# the XML output. Note that enabling this will significantly increase the size
+# of the XML output.
+# The default value is: YES.
+# This tag requires that the tag GENERATE_XML is set to YES.
+
+XML_PROGRAMLISTING     = YES
+
+#---------------------------------------------------------------------------
+# Configuration options related to the DOCBOOK output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_DOCBOOK tag is set to YES, doxygen will generate Docbook files
+# that can be used to generate PDF.
+# The default value is: NO.
+
+GENERATE_DOCBOOK       = NO
+
+# The DOCBOOK_OUTPUT tag is used to specify where the Docbook pages will be put.
+# If a relative path is entered the value of OUTPUT_DIRECTORY will be put in
+# front of it.
+# The default directory is: docbook.
+# This tag requires that the tag GENERATE_DOCBOOK is set to YES.
+
+DOCBOOK_OUTPUT         = docbook
+
+# If the DOCBOOK_PROGRAMLISTING tag is set to YES, doxygen will include the
+# program listings (including syntax highlighting and cross-referencing
+# information) to the DOCBOOK output. Note that enabling this will significantly
+# increase the size of the DOCBOOK output.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_DOCBOOK is set to YES.
+
+DOCBOOK_PROGRAMLISTING = NO
+
+#---------------------------------------------------------------------------
+# Configuration options for the AutoGen Definitions output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_AUTOGEN_DEF tag is set to YES, doxygen will generate an
+# AutoGen Definitions (see http://autogen.sf.net) file that captures the
+# structure of the code including all documentation. Note that this feature is
+# still experimental and incomplete at the moment.
+# The default value is: NO.
+
+GENERATE_AUTOGEN_DEF   = NO
+
+#---------------------------------------------------------------------------
+# Configuration options related to the Perl module output
+#---------------------------------------------------------------------------
+
+# If the GENERATE_PERLMOD tag is set to YES, doxygen will generate a Perl module
+# file that captures the structure of the code including all documentation.
+#
+# Note that this feature is still experimental and incomplete at the moment.
+# The default value is: NO.
+
+GENERATE_PERLMOD       = NO
+
+# If the PERLMOD_LATEX tag is set to YES, doxygen will generate the necessary
+# Makefile rules, Perl scripts and LaTeX code to be able to generate PDF and DVI
+# output from the Perl module output.
+# The default value is: NO.
+# This tag requires that the tag GENERATE_PERLMOD is set to YES.
+
+PERLMOD_LATEX          = NO
+
+# If the PERLMOD_PRETTY tag is set to YES, the Perl module output will be nicely
+# formatted so it can be parsed by a human reader. This is useful if you want to
+# understand what is going on. On the other hand, if this tag is set to NO, the
+# size of the Perl module output will be much smaller and Perl will parse it
+# just the same.
+# The default value is: YES.
+# This tag requires that the tag GENERATE_PERLMOD is set to YES.
+
+PERLMOD_PRETTY         = YES
+
+# The names of the make variables in the generated doxyrules.make file are
+# prefixed with the string contained in PERLMOD_MAKEVAR_PREFIX. This is useful
+# so different doxyrules.make files included by the same Makefile don't
+# overwrite each other's variables.
+# This tag requires that the tag GENERATE_PERLMOD is set to YES.
+
+PERLMOD_MAKEVAR_PREFIX =
+
+#---------------------------------------------------------------------------
+# Configuration options related to the preprocessor
+#---------------------------------------------------------------------------
+
+# If the ENABLE_PREPROCESSING tag is set to YES, doxygen will evaluate all
+# C-preprocessor directives found in the sources and include files.
+# The default value is: YES.
+
+ENABLE_PREPROCESSING   = YES
+
+# If the MACRO_EXPANSION tag is set to YES, doxygen will expand all macro names
+# in the source code. If set to NO, only conditional compilation will be
+# performed. Macro expansion can be done in a controlled way by setting
+# EXPAND_ONLY_PREDEF to YES.
+# The default value is: NO.
+# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
+
+MACRO_EXPANSION        = YES
+
+# If the EXPAND_ONLY_PREDEF and MACRO_EXPANSION tags are both set to YES then
+# the macro expansion is limited to the macros specified with the PREDEFINED and
+# EXPAND_AS_DEFINED tags.
+# The default value is: NO.
+# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
+
+EXPAND_ONLY_PREDEF     = YES
+
+# If the SEARCH_INCLUDES tag is set to YES, the include files in the
+# INCLUDE_PATH will be searched if a #include is found.
+# The default value is: YES.
+# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
+
+SEARCH_INCLUDES        = NO
+
+# The INCLUDE_PATH tag can be used to specify one or more directories that
+# contain include files that are not input files but should be processed by the
+# preprocessor.
+# This tag requires that the tag SEARCH_INCLUDES is set to YES.
+
+INCLUDE_PATH           =
+
+# You can use the INCLUDE_FILE_PATTERNS tag to specify one or more wildcard
+# patterns (like *.h and *.hpp) to filter out the header-files in the
+# directories. If left blank, the patterns specified with FILE_PATTERNS will be
+# used.
+# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
+
+INCLUDE_FILE_PATTERNS  =
+
+# The PREDEFINED tag can be used to specify one or more macro names that are
+# defined before the preprocessor is started (similar to the -D option of e.g.
+# gcc). The argument of the tag is a list of macros of the form: name or
+# name=definition (no spaces). If the definition and the "=" are omitted, "=1"
+# is assumed. To prevent a macro definition from being undefined via #undef or
+# recursively expanded use the := operator instead of the = operator.
+# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
+
+PREDEFINED             = __attribute__(x)= \
+                         __inline=
+
+# If the MACRO_EXPANSION and EXPAND_ONLY_PREDEF tags are set to YES then this
+# tag can be used to specify a list of macro names that should be expanded. The
+# macro definition that is found in the sources will be used. Use the PREDEFINED
+# tag if you want to use a different macro definition that overrules the
+# definition found in the source code.
+# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
+
+EXPAND_AS_DEFINED      =
+
+# If the SKIP_FUNCTION_MACROS tag is set to YES then doxygen's preprocessor will
+# remove all references to function-like macros that are alone on a line, have
+# an all uppercase name, and do not end with a semicolon. Such function macros
+# are typically used for boiler-plate code, and will confuse the parser if not
+# removed.
+# The default value is: YES.
+# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
+
+SKIP_FUNCTION_MACROS   = YES
+
+#---------------------------------------------------------------------------
+# Configuration options related to external references
+#---------------------------------------------------------------------------
+
+# The TAGFILES tag can be used to specify one or more tag files. For each tag
+# file the location of the external documentation should be added. The format of
+# a tag file without this location is as follows:
+# TAGFILES = file1 file2 ...
+# Adding location for the tag files is done as follows:
+# TAGFILES = file1=loc1 "file2 = loc2" ...
+# where loc1 and loc2 can be relative or absolute paths or URLs. See the
+# section "Linking to external documentation" for more information about the use
+# of tag files.
+# Note: Each tag file must have a unique name (where the name does NOT include
+# the path). If a tag file is not located in the directory in which doxygen is
+# run, you must also specify the path to the tagfile here.
+
+TAGFILES               =
+
+# When a file name is specified after GENERATE_TAGFILE, doxygen will create a
+# tag file that is based on the input files it reads. See section "Linking to
+# external documentation" for more information about the usage of tag files.
+
+GENERATE_TAGFILE       =
+
+# If the ALLEXTERNALS tag is set to YES, all external class will be listed in
+# the class index. If set to NO, only the inherited external classes will be
+# listed.
+# The default value is: NO.
+
+ALLEXTERNALS           = NO
+
+# If the EXTERNAL_GROUPS tag is set to YES, all external groups will be listed
+# in the modules index. If set to NO, only the current project's groups will be
+# listed.
+# The default value is: YES.
+
+EXTERNAL_GROUPS        = YES
+
+# If the EXTERNAL_PAGES tag is set to YES, all external pages will be listed in
+# the related pages index. If set to NO, only the current project's pages will
+# be listed.
+# The default value is: YES.
+
+EXTERNAL_PAGES         = YES
+
+# The PERL_PATH should be the absolute path and name of the perl script
+# interpreter (i.e. the result of 'which perl').
+# The default file (with absolute path) is: /usr/bin/perl.
+
+PERL_PATH              = /usr/bin/perl
+
+#---------------------------------------------------------------------------
+# Configuration options related to the dot tool
+#---------------------------------------------------------------------------
+
+# If the CLASS_DIAGRAMS tag is set to YES, doxygen will generate a class diagram
+# (in HTML and LaTeX) for classes with base or super classes. Setting the tag to
+# NO turns the diagrams off. Note that this option also works with HAVE_DOT
+# disabled, but it is recommended to install and use dot, since it yields more
+# powerful graphs.
+# The default value is: YES.
+
+CLASS_DIAGRAMS         = NO
+
+# You can define message sequence charts within doxygen comments using the \msc
+# command. Doxygen will then run the mscgen tool (see:
+# http://www.mcternan.me.uk/mscgen/)) to produce the chart and insert it in the
+# documentation. The MSCGEN_PATH tag allows you to specify the directory where
+# the mscgen tool resides. If left empty the tool is assumed to be found in the
+# default search path.
+
+MSCGEN_PATH            =
+
+# You can include diagrams made with dia in doxygen documentation. Doxygen will
+# then run dia to produce the diagram and insert it in the documentation. The
+# DIA_PATH tag allows you to specify the directory where the dia binary resides.
+# If left empty dia is assumed to be found in the default search path.
+
+DIA_PATH               =
+
+# If set to YES the inheritance and collaboration graphs will hide inheritance
+# and usage relations if the target is undocumented or is not a class.
+# The default value is: YES.
+
+HIDE_UNDOC_RELATIONS   = YES
+
+# If you set the HAVE_DOT tag to YES then doxygen will assume the dot tool is
+# available from the path. This tool is part of Graphviz (see:
+# http://www.graphviz.org/), a graph visualization toolkit from AT&T and Lucent
+# Bell Labs. The other options in this section have no effect if this option is
+# set to NO
+# The default value is: NO.
+
+HAVE_DOT               = NO
+
+# The DOT_NUM_THREADS specifies the number of dot invocations doxygen is allowed
+# to run in parallel. When set to 0 doxygen will base this on the number of
+# processors available in the system. You can set it explicitly to a value
+# larger than 0 to get control over the balance between CPU load and processing
+# speed.
+# Minimum value: 0, maximum value: 32, default value: 0.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_NUM_THREADS        = 0
+
+# When you want a differently looking font in the dot files that doxygen
+# generates you can specify the font name using DOT_FONTNAME. You need to make
+# sure dot is able to find the font, which can be done by putting it in a
+# standard location or by setting the DOTFONTPATH environment variable or by
+# setting DOT_FONTPATH to the directory containing the font.
+# The default value is: Helvetica.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_FONTNAME           = Helvetica
+
+# The DOT_FONTSIZE tag can be used to set the size (in points) of the font of
+# dot graphs.
+# Minimum value: 4, maximum value: 24, default value: 10.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_FONTSIZE           = 10
+
+# By default doxygen will tell dot to use the default font as specified with
+# DOT_FONTNAME. If you specify a different font using DOT_FONTNAME you can set
+# the path where dot can find it using this tag.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_FONTPATH           =
+
+# If the CLASS_GRAPH tag is set to YES then doxygen will generate a graph for
+# each documented class showing the direct and indirect inheritance relations.
+# Setting this tag to YES will force the CLASS_DIAGRAMS tag to NO.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+CLASS_GRAPH            = YES
+
+# If the COLLABORATION_GRAPH tag is set to YES then doxygen will generate a
+# graph for each documented class showing the direct and indirect implementation
+# dependencies (inheritance, containment, and class references variables) of the
+# class with other documented classes.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+COLLABORATION_GRAPH    = YES
+
+# If the GROUP_GRAPHS tag is set to YES then doxygen will generate a graph for
+# groups, showing the direct groups dependencies.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+GROUP_GRAPHS           = YES
+
+# If the UML_LOOK tag is set to YES, doxygen will generate inheritance and
+# collaboration diagrams in a style similar to the OMG's Unified Modeling
+# Language.
+# The default value is: NO.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+UML_LOOK               = NO
+
+# If the UML_LOOK tag is enabled, the fields and methods are shown inside the
+# class node. If there are many fields or methods and many nodes the graph may
+# become too big to be useful. The UML_LIMIT_NUM_FIELDS threshold limits the
+# number of items for each type to make the size more manageable. Set this to 0
+# for no limit. Note that the threshold may be exceeded by 50% before the limit
+# is enforced. So when you set the threshold to 10, up to 15 fields may appear,
+# but if the number exceeds 15, the total amount of fields shown is limited to
+# 10.
+# Minimum value: 0, maximum value: 100, default value: 10.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+UML_LIMIT_NUM_FIELDS   = 10
+
+# If the TEMPLATE_RELATIONS tag is set to YES then the inheritance and
+# collaboration graphs will show the relations between templates and their
+# instances.
+# The default value is: NO.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+TEMPLATE_RELATIONS     = NO
+
+# If the INCLUDE_GRAPH, ENABLE_PREPROCESSING and SEARCH_INCLUDES tags are set to
+# YES then doxygen will generate a graph for each documented file showing the
+# direct and indirect include dependencies of the file with other documented
+# files.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+INCLUDE_GRAPH          = YES
+
+# If the INCLUDED_BY_GRAPH, ENABLE_PREPROCESSING and SEARCH_INCLUDES tags are
+# set to YES then doxygen will generate a graph for each documented file showing
+# the direct and indirect include dependencies of the file with other documented
+# files.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+INCLUDED_BY_GRAPH      = YES
+
+# If the CALL_GRAPH tag is set to YES then doxygen will generate a call
+# dependency graph for every global function or class method.
+#
+# Note that enabling this option will significantly increase the time of a run.
+# So in most cases it will be better to enable call graphs for selected
+# functions only using the \callgraph command. Disabling a call graph can be
+# accomplished by means of the command \hidecallgraph.
+# The default value is: NO.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+CALL_GRAPH             = NO
+
+# If the CALLER_GRAPH tag is set to YES then doxygen will generate a caller
+# dependency graph for every global function or class method.
+#
+# Note that enabling this option will significantly increase the time of a run.
+# So in most cases it will be better to enable caller graphs for selected
+# functions only using the \callergraph command. Disabling a caller graph can be
+# accomplished by means of the command \hidecallergraph.
+# The default value is: NO.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+CALLER_GRAPH           = NO
+
+# If the GRAPHICAL_HIERARCHY tag is set to YES then doxygen will graphical
+# hierarchy of all classes instead of a textual one.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+GRAPHICAL_HIERARCHY    = YES
+
+# If the DIRECTORY_GRAPH tag is set to YES then doxygen will show the
+# dependencies a directory has on other directories in a graphical way. The
+# dependency relations are determined by the #include relations between the
+# files in the directories.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DIRECTORY_GRAPH        = YES
+
+# The DOT_IMAGE_FORMAT tag can be used to set the image format of the images
+# generated by dot. For an explanation of the image formats see the section
+# output formats in the documentation of the dot tool (Graphviz (see:
+# http://www.graphviz.org/)).
+# Note: If you choose svg you need to set HTML_FILE_EXTENSION to xhtml in order
+# to make the SVG files visible in IE 9+ (other browsers do not have this
+# requirement).
+# Possible values are: png, jpg, gif, svg, png:gd, png:gd:gd, png:cairo,
+# png:cairo:gd, png:cairo:cairo, png:cairo:gdiplus, png:gdiplus and
+# png:gdiplus:gdiplus.
+# The default value is: png.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_IMAGE_FORMAT       = png
+
+# If DOT_IMAGE_FORMAT is set to svg, then this option can be set to YES to
+# enable generation of interactive SVG images that allow zooming and panning.
+#
+# Note that this requires a modern browser other than Internet Explorer. Tested
+# and working are Firefox, Chrome, Safari, and Opera.
+# Note: For IE 9+ you need to set HTML_FILE_EXTENSION to xhtml in order to make
+# the SVG files visible. Older versions of IE do not have SVG support.
+# The default value is: NO.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+INTERACTIVE_SVG        = NO
+
+# The DOT_PATH tag can be used to specify the path where the dot tool can be
+# found. If left blank, it is assumed the dot tool can be found in the path.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_PATH               =
+
+# The DOTFILE_DIRS tag can be used to specify one or more directories that
+# contain dot files that are included in the documentation (see the \dotfile
+# command).
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOTFILE_DIRS           =
+
+# The MSCFILE_DIRS tag can be used to specify one or more directories that
+# contain msc files that are included in the documentation (see the \mscfile
+# command).
+
+MSCFILE_DIRS           =
+
+# The DIAFILE_DIRS tag can be used to specify one or more directories that
+# contain dia files that are included in the documentation (see the \diafile
+# command).
+
+DIAFILE_DIRS           =
+
+# When using plantuml, the PLANTUML_JAR_PATH tag should be used to specify the
+# path where java can find the plantuml.jar file. If left blank, it is assumed
+# PlantUML is not used or called during a preprocessing step. Doxygen will
+# generate a warning when it encounters a \startuml command in this case and
+# will not generate output for the diagram.
+
+PLANTUML_JAR_PATH      =
+
+# When using plantuml, the specified paths are searched for files specified by
+# the !include statement in a plantuml block.
+
+PLANTUML_INCLUDE_PATH  =
+
+# The DOT_GRAPH_MAX_NODES tag can be used to set the maximum number of nodes
+# that will be shown in the graph. If the number of nodes in a graph becomes
+# larger than this value, doxygen will truncate the graph, which is visualized
+# by representing a node as a red box. Note that doxygen if the number of direct
+# children of the root node in a graph is already larger than
+# DOT_GRAPH_MAX_NODES then the graph will not be shown at all. Also note that
+# the size of a graph can be further restricted by MAX_DOT_GRAPH_DEPTH.
+# Minimum value: 0, maximum value: 10000, default value: 50.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_GRAPH_MAX_NODES    = 50
+
+# The MAX_DOT_GRAPH_DEPTH tag can be used to set the maximum depth of the graphs
+# generated by dot. A depth value of 3 means that only nodes reachable from the
+# root by following a path via at most 3 edges will be shown. Nodes that lay
+# further from the root node will be omitted. Note that setting this option to 1
+# or 2 may greatly reduce the computation time needed for large code bases. Also
+# note that the size of a graph can be further restricted by
+# DOT_GRAPH_MAX_NODES. Using a depth of 0 means no depth restriction.
+# Minimum value: 0, maximum value: 1000, default value: 0.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+MAX_DOT_GRAPH_DEPTH    = 0
+
+# Set the DOT_TRANSPARENT tag to YES to generate images with a transparent
+# background. This is disabled by default, because dot on Windows does not seem
+# to support this out of the box.
+#
+# Warning: Depending on the platform used, enabling this option may lead to
+# badly anti-aliased labels on the edges of a graph (i.e. they become hard to
+# read).
+# The default value is: NO.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_TRANSPARENT        = NO
+
+# Set the DOT_MULTI_TARGETS tag to YES to allow dot to generate multiple output
+# files in one run (i.e. multiple -o and -T options on the command line). This
+# makes dot run faster, but since only newer versions of dot (>1.8.10) support
+# this, this feature is disabled by default.
+# The default value is: NO.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_MULTI_TARGETS      = NO
+
+# If the GENERATE_LEGEND tag is set to YES doxygen will generate a legend page
+# explaining the meaning of the various boxes and arrows in the dot generated
+# graphs.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+GENERATE_LEGEND        = YES
+
+# If the DOT_CLEANUP tag is set to YES, doxygen will remove the intermediate dot
+# files that are used to generate the various graphs.
+# The default value is: YES.
+# This tag requires that the tag HAVE_DOT is set to YES.
+
+DOT_CLEANUP            = YES
diff --git a/docs/index.rst b/docs/index.rst
new file mode 100644
index 0000000000000000000000000000000000000000..f4e66c1b51f4528fa47057116a8f457680b3d972
--- /dev/null
+++ b/docs/index.rst
@@ -0,0 +1,52 @@
+============================
+Composable Kernel User Guide
+============================
+
+------------
+Introduction
+------------
+
+This document contains instructions for installing, using, and contributing to Composable Kernel (CK).
+
+-----------
+Methodology
+-----------
+
+Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
+
+CK utilizes two concepts to achieve performance portability and code maintainability:
+
+* A tile-based programming model
+* Algorithm complexity reduction for complex ML operators, using innovative technique we call "Tensor Coordinate Transformation".
+
+.. image:: data/ck_component.png
+   :alt: CK Components
+
+--------------
+Code Structure
+--------------
+
+Current CK library are structured into 4 layers:
+
+* "Templated Tile Operators" layer
+* "Templated Kernel and Invoker" layer
+* "Instantiated Kernel and Invoker" layer
+* "Client API" layer
+
+.. image:: data/ck_layer.png
+   :alt: CK Layers
+   
+Documentation Roadmap
+^^^^^^^^^^^^^^^^^^^^^
+The following is a list of CK documents in the suggested reading order:
+
+.. toctree::
+   :maxdepth: 5
+   :caption: Contents:
+   :numbered:
+
+   tutorial_hello_world
+   dockerhub
+   Supported_Primitives_Guide
+   API_Reference_Guide
+   Contributors_Guide
diff --git a/docs/license.rst b/docs/license.rst
new file mode 100644
index 0000000000000000000000000000000000000000..ddb544496e41b92dcaadb6ff816946e46ca16da8
--- /dev/null
+++ b/docs/license.rst
@@ -0,0 +1,6 @@
+=======
+License
+=======
+
+.. include:: ../LICENSE
+   :literal:
diff --git a/docs/refs.bib b/docs/refs.bib
new file mode 100644
index 0000000000000000000000000000000000000000..3c6d775a7e62fb72cdd7491da3f46270e95566eb
--- /dev/null
+++ b/docs/refs.bib
@@ -0,0 +1,7 @@
+
+@article{dao2022flashattention,
+  title={Flashattention: Fast and memory-efficient exact attention with io-awareness},
+  author={Dao, Tri and Fu, Daniel Y and Ermon, Stefano and Rudra, Atri and R{\'e}, Christopher},
+  journal={arXiv preprint arXiv:2205.14135},
+  year={2022}
+}
diff --git a/docs/sphinx/_toc.yml.in b/docs/sphinx/_toc.yml.in
new file mode 100644
index 0000000000000000000000000000000000000000..83dd1e7b1a5d759143f2db13a1e296f5e66da96d
--- /dev/null
+++ b/docs/sphinx/_toc.yml.in
@@ -0,0 +1,10 @@
+# Anywhere {branch} is used, the branch name will be substituted.
+# These comments will also be removed.
+defaults:
+  numbered: False
+  maxdepth: 6
+root: index
+subtrees:
+  - caption: About
+    entries:
+      - file: license
diff --git a/docs/sphinx/requirements.in b/docs/sphinx/requirements.in
new file mode 100644
index 0000000000000000000000000000000000000000..4bdf41b959975f9b7f6ba38d20d5070f83e4da1e
--- /dev/null
+++ b/docs/sphinx/requirements.in
@@ -0,0 +1,2 @@
+rocm-docs-core==0.10.3
+sphinxcontrib-bibtex==2.5.0
diff --git a/docs/sphinx/requirements.txt b/docs/sphinx/requirements.txt
new file mode 100644
index 0000000000000000000000000000000000000000..097acba2257d3c3f1c03d801ebeebdccd0dd120f
--- /dev/null
+++ b/docs/sphinx/requirements.txt
@@ -0,0 +1,172 @@
+#
+# This file is autogenerated by pip-compile with Python 3.8
+# by the following command:
+#
+#    pip-compile requirements.in
+#
+accessible-pygments==0.0.3
+    # via pydata-sphinx-theme
+alabaster==0.7.13
+    # via sphinx
+babel==2.12.1
+    # via
+    #   pydata-sphinx-theme
+    #   sphinx
+beautifulsoup4==4.11.2
+    # via pydata-sphinx-theme
+breathe==4.34.0
+    # via rocm-docs-core
+certifi==2022.12.7
+    # via requests
+cffi==1.15.1
+    # via
+    #   cryptography
+    #   pynacl
+charset-normalizer==3.1.0
+    # via requests
+click==8.1.3
+    # via sphinx-external-toc
+cryptography==40.0.2
+    # via pyjwt
+deprecated==1.2.13
+    # via pygithub
+docutils==0.16
+    # via
+    #   breathe
+    #   myst-parser
+    #   pybtex-docutils
+    #   pydata-sphinx-theme
+    #   sphinx
+    #   sphinxcontrib-bibtex
+gitdb==4.0.10
+    # via gitpython
+gitpython==3.1.31
+    # via rocm-docs-core
+idna==3.4
+    # via requests
+imagesize==1.4.1
+    # via sphinx
+importlib-metadata==6.0.0
+    # via
+    #   sphinx
+    #   sphinxcontrib-bibtex
+importlib-resources==5.12.0
+    # via rocm-docs-core
+jinja2==3.1.2
+    # via
+    #   myst-parser
+    #   sphinx
+latexcodec==2.0.1
+    # via pybtex
+linkify-it-py==1.0.3
+    # via myst-parser
+markdown-it-py==2.2.0
+    # via
+    #   mdit-py-plugins
+    #   myst-parser
+markupsafe==2.1.2
+    # via jinja2
+mdit-py-plugins==0.3.5
+    # via myst-parser
+mdurl==0.1.2
+    # via markdown-it-py
+myst-parser[linkify]==1.0.0
+    # via rocm-docs-core
+packaging==23.0
+    # via
+    #   pydata-sphinx-theme
+    #   sphinx
+pybtex==0.24.0
+    # via
+    #   pybtex-docutils
+    #   sphinxcontrib-bibtex
+pybtex-docutils==1.0.2
+    # via sphinxcontrib-bibtex
+pycparser==2.21
+    # via cffi
+pydata-sphinx-theme==0.13.3
+    # via
+    #   rocm-docs-core
+    #   sphinx-book-theme
+pygithub==1.58.2
+    # via rocm-docs-core
+pygments==2.14.0
+    # via
+    #   accessible-pygments
+    #   pydata-sphinx-theme
+    #   sphinx
+pyjwt[crypto]==2.6.0
+    # via pygithub
+pynacl==1.5.0
+    # via pygithub
+pytz==2023.3
+    # via babel
+pyyaml==6.0
+    # via
+    #   myst-parser
+    #   pybtex
+    #   sphinx-external-toc
+requests==2.28.2
+    # via
+    #   pygithub
+    #   sphinx
+rocm-docs-core==0.10.3
+    # via -r requirements.in
+six==1.16.0
+    # via
+    #   latexcodec
+    #   pybtex
+smmap==5.0.0
+    # via gitdb
+snowballstemmer==2.2.0
+    # via sphinx
+soupsieve==2.4
+    # via beautifulsoup4
+sphinx==5.3.0
+    # via
+    #   breathe
+    #   myst-parser
+    #   pydata-sphinx-theme
+    #   rocm-docs-core
+    #   sphinx-book-theme
+    #   sphinx-copybutton
+    #   sphinx-design
+    #   sphinx-external-toc
+    #   sphinx-notfound-page
+    #   sphinxcontrib-bibtex
+sphinx-book-theme==1.0.1
+    # via rocm-docs-core
+sphinx-copybutton==0.5.1
+    # via rocm-docs-core
+sphinx-design==0.3.0
+    # via rocm-docs-core
+sphinx-external-toc==0.3.1
+    # via rocm-docs-core
+sphinx-notfound-page==0.8.3
+    # via rocm-docs-core
+sphinxcontrib-applehelp==1.0.4
+    # via sphinx
+sphinxcontrib-bibtex==2.5.0
+    # via -r requirements.in
+sphinxcontrib-devhelp==1.0.2
+    # via sphinx
+sphinxcontrib-htmlhelp==2.0.1
+    # via sphinx
+sphinxcontrib-jsmath==1.0.1
+    # via sphinx
+sphinxcontrib-qthelp==1.0.3
+    # via sphinx
+sphinxcontrib-serializinghtml==1.1.5
+    # via sphinx
+typing-extensions==4.5.0
+    # via pydata-sphinx-theme
+uc-micro-py==1.0.1
+    # via linkify-it-py
+urllib3==1.26.15
+    # via requests
+wrapt==1.15.0
+    # via deprecated
+zipp==3.15.0
+    # via
+    #   importlib-metadata
+    #   importlib-resources
diff --git a/docs/tutorial_hello_world.rst b/docs/tutorial_hello_world.rst
new file mode 100644
index 0000000000000000000000000000000000000000..b8fd094654cba671b5082d8052b7dd3d6671867a
--- /dev/null
+++ b/docs/tutorial_hello_world.rst
@@ -0,0 +1,174 @@
+===============
+CK Hello world
+===============
+
+-------------------------------------
+Motivation
+-------------------------------------
+
+This tutorial is aimed at engineers dealing with artificial intelligence and machine learning who would like to optimize their pipelines and squeeze every performance drop by adding Composable Kernel (CK) library to their projects. We would like to make the CK library approachable so the tutorial is not based on the latest release and doesn't have all the bleeding edge features, but it will be reproducible now and forever.
+
+During this tutorial we will have an introduction to the CK library, we will build it and run some examples and tests, so to say we will run a "Hello world" example. In future tutorials we will go in depth and breadth and get familiar with other tools and ways to integrate CK into your project.
+
+-------------------------------------
+Description
+-------------------------------------
+
+Modern AI technology solves more and more problems in all imaginable fields, but crafting fast and efficient workflows is still challenging. CK is one of the tools to make AI heavy lifting as fast and efficient as possible. CK is a collection of optimized AI operator kernels and tools to create new ones. The library has components required for majority of modern neural networks architectures including matrix multiplication, convolution, contraction, reduction, attention modules, variety of activation functions, fused operators and many more.
+
+So how do we (almost) reach the speed of light? CK acceleration abilities are based on:
+
+* Layered structure.
+* Tile-based computation model.
+* Tensor coordinate transformation.
+* Hardware acceleration use.
+* Support of low precision data types including fp16, bf16, int8 and int4.
+
+If you are excited and need more technical details and benchmarking results - read this awesome `blog post <https://community.amd.com/t5/instinct-accelerators/amd-composable-kernel-library-efficient-fused-kernels-for-ai/ba-p/553224>`_.
+
+For more details visit our `github repo <https://github.com/ROCmSoftwarePlatform/composable_kernel>`_.
+
+-------------------------------------
+Hardware targets
+-------------------------------------
+
+CK library fully supports "gfx908" and "gfx90a" GPU architectures and only some operators are supported for "gfx1030". Let's check the hardware you have at hand and decide on the target GPU architecture
+
+==========     =========
+GPU Target     AMD GPU
+==========     =========
+gfx908 	       Radeon Instinct MI100
+gfx90a 	       Radeon Instinct MI210, MI250, MI250X
+gfx1030        Radeon PRO V620, W6800, W6800X, W6800X Duo, W6900X, RX 6800, RX 6800 XT, RX 6900 XT, RX 6900 XTX, RX 6950 XT
+==========     =========
+
+There are also `cloud options <https://aws.amazon.com/ec2/instance-types/g4/>`_ you can find if you don't have an AMD GPU at hand.
+
+-------------------------------------
+Build the library
+-------------------------------------
+
+First let's clone the library and rebase to the tested version::
+
+    git clone https://github.com/ROCmSoftwarePlatform/composable_kernel.git
+    cd composable_kernel/
+    git checkout tutorial_hello_world
+
+To make our lives easier we prepared `docker images <https://hub.docker.com/r/rocm/composable_kernel>`_ with all the necessary dependencies. Pick the right image and create a container. In this tutorial we use "rocm/composable_kernel:ck_ub20.04_rocm5.3_release" image, it is based on Ubuntu 20.04, ROCm v5.3, compiler release version.
+
+If your current folder is ${HOME}, start the docker container with::
+
+    docker run  \
+    -it  \
+    --privileged  \
+    --group-add sudo  \
+    -w /root/workspace  \
+    -v ${HOME}:/root/workspace  \
+    rocm/composable_kernel:ck_ub20.04_rocm5.3_release  \
+    /bin/bash
+
+If your current folder is different from ${HOME}, adjust the line `-v ${HOME}:/root/workspace` to fit your folder structure.
+
+Inside the docker container current folder is "~/workspace", library path is "~/workspace/composable_kernel", navigate to the library::
+
+    cd composable_kernel/
+
+Create and go to the "build" directory::
+
+    mkdir build && cd build
+
+In the previous section we talked about target GPU architecture. Once you decide which one is right for you, run cmake using the right GPU_TARGETS flag::
+
+    cmake  \
+    -D CMAKE_PREFIX_PATH=/opt/rocm  \
+    -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc  \
+    -D CMAKE_CXX_FLAGS="-O3"  \
+    -D CMAKE_BUILD_TYPE=Release  \
+    -D BUILD_DEV=OFF  \
+    -D GPU_TARGETS="gfx908;gfx90a;gfx1030" ..
+
+If everything went well the cmake run will end up with::
+
+    -- Configuring done
+    -- Generating done
+    -- Build files have been written to: "/root/workspace/composable_kernel/build"
+
+Finally, we can build examples and tests::
+
+    make -j examples tests
+
+If everything is smooth, you'll see::
+
+    Scanning dependencies of target tests
+    [100%] Built target tests
+
+---------------------------
+Run examples and tests
+---------------------------
+
+Examples are listed as test cases as well, so we can run all examples and tests with::
+
+    ctest
+
+You can check the list of all tests by running::
+
+    ctest -N
+
+We can also run them separately, here is a separate example execution::
+
+    ./bin/example_gemm_xdl_fp16 1 1 1
+
+The arguments "1 1 1" mean that we want to run this example in the mode: verify results with CPU, initialize matrices with integers and benchmark the kernel execution. You can play around with these parameters and see how output and execution results change.
+
+If everything goes well and you have a device based on gfx908 or gfx90a architecture you should see something like::
+
+    a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
+    b_k_n: dim 2, lengths {4096, 4096}, strides {1, 4096}
+    c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
+    launch_and_time_kernel: grid_dim {480, 1, 1}, block_dim {256, 1, 1}
+    Warm up 1 time
+    Start running 10 times...
+    Perf: 1.10017 ms, 117.117 TFlops, 87.6854 GB/s, DeviceGemmXdl<256, 256, 128, 4, 8, 32, 32, 4, 2> NumPrefetch: 1, LoopScheduler: Default, PipelineVersion: v1
+
+Meanwhile, running it on a gfx1030 device should result in::
+
+    a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
+    b_k_n: dim 2, lengths {4096, 4096}, strides {1, 4096}
+    c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
+    DeviceGemmXdl<256, 256, 128, 4, 8, 32, 32, 4, 2> NumPrefetch: 1, LoopScheduler: Default, PipelineVersion: v1 does not support this problem
+
+But don't panic, some of the operators are supported on gfx1030 architecture, so you can run a separate example like::
+
+    ./bin/example_gemm_dl_fp16 1 1 1
+
+and it should result in something nice similar to::
+
+    a_m_k: dim 2, lengths {3840, 4096}, strides {1, 4096}
+    b_k_n: dim 2, lengths {4096, 4096}, strides {4096, 1}
+    c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
+    arg.a_grid_desc_k0_m0_m1_k1_{2048, 3840, 2}
+    arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
+    arg.c_grid_desc_m_n_{ 3840, 4096}
+    launch_and_time_kernel: grid_dim {960, 1, 1}, block_dim {256, 1, 1}
+    Warm up 1 time
+    Start running 10 times...
+    Perf: 3.65695 ms, 35.234 TFlops, 26.3797 GB/s, DeviceGemmDl<256, 128, 128, 16, 2, 4, 4, 1>
+
+Or we can run a separate test::
+
+    ctest -R test_gemm_fp16
+
+If everything goes well you should see something like::
+
+    Start 121: test_gemm_fp16
+    1/1 Test #121: test_gemm_fp16 ...................   Passed   51.81 sec
+
+    100% tests passed, 0 tests failed out of 1
+
+-----------
+Summary
+-----------
+
+In this tutorial we took the first look at the Composable Kernel library, built it on your system and ran some examples and tests. Stay tuned, in the next tutorial we will run kernels with different configs to find out the best one for your hardware and task.
+
+P.S.: Don't forget to switch out the cloud instance if you have launched one, you can find better ways to spend your money for sure!
diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index c403e51ed99d10cef3ecf5bc631f3f20fa48a1ab..66afe73c2a970a92934431d191659dfee31c19b2 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -17,12 +17,14 @@ endif(USE_BITINT_EXTENSION_INT4)
 add_custom_target(example_gemm_xdl)
 
 add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
+add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
 add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
 add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
 
 add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
 add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
 add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
+add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
 
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_gemm_xdl_int4 gemm_xdl_int4.cpp)
@@ -35,3 +37,14 @@ add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
 
 add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
 add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
+
+if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+  add_custom_target(example_gemm_wmma)
+  add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
+  add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
+endif()
+
+if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+  add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
+endif()
diff --git a/example/01_gemm/common.hpp b/example/01_gemm/common.hpp
index 495a8159623beb22d6002bb2e1667ef459b74139..144c9aaccd7f8b1e20e30e671be4be2525ee1d03 100644
--- a/example/01_gemm/common.hpp
+++ b/example/01_gemm/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/01_gemm/gemm_dl_fp16.cpp b/example/01_gemm/gemm_dl_fp16.cpp
index cf585a8c51cbb9b5b0218228e4aa706189598804..b5fecb97521bd2d4213a9435b01dafe45a43be6e 100644
--- a/example/01_gemm/gemm_dl_fp16.cpp
+++ b/example/01_gemm/gemm_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_dl_fp32.cpp b/example/01_gemm/gemm_dl_fp32.cpp
index 93f085cdee53667a5b906cfed3b037d57d00bf5f..212b72f2a6a060a602c5c640709da77684c7c55e 100644
--- a/example/01_gemm/gemm_dl_fp32.cpp
+++ b/example/01_gemm/gemm_dl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_dl_int4.cpp b/example/01_gemm/gemm_dl_int4.cpp
index e392c490f29a48da3a6424f46876e220306f2907..e55ae140130c0779c1e81e0bdc84b6724c6bac86 100644
--- a/example/01_gemm/gemm_dl_int4.cpp
+++ b/example/01_gemm/gemm_dl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/01_gemm/gemm_dl_int8.cpp b/example/01_gemm/gemm_dl_int8.cpp
index be9e387718f120fb1ba708d374f0ff9a09fc806d..1840390aa9e02e85c88328baf358546db362ab24 100644
--- a/example/01_gemm/gemm_dl_int8.cpp
+++ b/example/01_gemm/gemm_dl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_wmma_fp16.cpp b/example/01_gemm/gemm_wmma_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b11fe76ab2ce031708be7addd7f84be0f15adc6b
--- /dev/null
+++ b/example/01_gemm/gemm_wmma_fp16.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp"
+
+using ADataType        = ck::half_t;
+using BDataType        = ck::half_t;
+using AccDataType      = float;
+using CShuffleDataType = float;
+using CDataType        = ck::half_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|              GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer|MRepeat|NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise|    Spacialization|  Size| Block| Block| Block|    | WMMA| WMMA|       |       |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN|MWmmaPerWave|NWmmaPerWave|        _MBlock_MWaveMPerWmma| ScalarPerVector|
+// ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|                  |      |      |      |      |    |     |     |       |       | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|        _NBlock_NWaveNPerWmma|  _NWaveNPerWmma|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |                  |      |      |      |      |    |     |     |       |       |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmMNKPadding,   256,   128,   256,     8,   8,   16,   16,      4,      4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,           1,           1,              S<1, 32, 1,  8>,               8, 1>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/example/01_gemm/gemm_xdl_bf16.cpp b/example/01_gemm/gemm_xdl_bf16.cpp
index 9aaae6ade9564aa54bc4ea0c7f2c96aac89ff505..3cac55ef4702856dd08dc223c02ea25113dbbf32 100644
--- a/example/01_gemm/gemm_xdl_bf16.cpp
+++ b/example/01_gemm/gemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_f8.cpp b/example/01_gemm/gemm_xdl_f8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..10159267776f6f33469affc618192e29d62763ac
--- /dev/null
+++ b/example/01_gemm/gemm_xdl_f8.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::f8_t;
+using BDataType        = ck::f8_t;
+using CDataType        = ck::f8_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::f8_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+// ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,              16>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/example/01_gemm/gemm_xdl_fp16.cpp b/example/01_gemm/gemm_xdl_fp16.cpp
index 488babb7588df3f754e7ba1c56eb739a24ac73b5..54fbd9cdd4983c1a810f3c8a1ac1521b363af690 100644
--- a/example/01_gemm/gemm_xdl_fp16.cpp
+++ b/example/01_gemm/gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
@@ -12,6 +12,8 @@ using AccDataType      = float;
 using CShuffleDataType = float;
 using CDataType        = ck::half_t;
 
+using F16 = ck::half_t;
+
 using ALayout = Row;
 using BLayout = Col;
 using CLayout = Row;
@@ -29,7 +31,7 @@ using DeviceGemmInstance0 = ck::tensor_operation::device::DeviceGemmXdl
 // ######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
 // ######|          |          |          |            |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
          < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1>;
-// clang-format on
+// // clang-format on
 
 // clang-format off
 using DeviceGemmInstance1 = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
@@ -40,7 +42,7 @@ using DeviceGemmInstance1 = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffl
          < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
 // clang-format on
 
-using DeviceGemmInstance = DeviceGemmInstance0;
+using DeviceGemmInstance = DeviceGemmInstance1;
 
 using ReferenceGemmInstance = ck::tensor_operation::host::
     ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
diff --git a/example/01_gemm/gemm_xdl_fp64.cpp b/example/01_gemm/gemm_xdl_fp64.cpp
index 99253b743d58707707c9765c18e5933cb09e4220..8361576299c3da3727cc467387e9717fceb93ce3 100644
--- a/example/01_gemm/gemm_xdl_fp64.cpp
+++ b/example/01_gemm/gemm_xdl_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_int4.cpp b/example/01_gemm/gemm_xdl_int4.cpp
index 7f1283a47b36c94be9645abd9fbd63094293f46e..f6238c7aa5040d8e409a139fe8c144835282cdc9 100644
--- a/example/01_gemm/gemm_xdl_int4.cpp
+++ b/example/01_gemm/gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/01_gemm/gemm_xdl_int8.cpp b/example/01_gemm/gemm_xdl_int8.cpp
index e67594c5bcbd601ae4747ac7720dab66b605ba2c..cc03200b9d153e8e22e13e6c77a52ad76ed79174 100644
--- a/example/01_gemm/gemm_xdl_int8.cpp
+++ b/example/01_gemm/gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp b/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
index 12a69925977b7eef2e5aa178016a7d433e9c0781..3afd0ebdb955dbcf949cfa2467a85cdf325c5f4f 100644
--- a/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
+++ b/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_wavelet_fp16.cpp b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b0f963fee51c20b089811ba2b6396b9d7125d8f6
--- /dev/null
+++ b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+
+using ADataType        = ck::half_t;
+using BDataType        = ck::half_t;
+using AccDataType      = float;
+using CShuffleDataType = float;
+using CDataType        = ck::half_t;
+
+using F16 = ck::half_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_WaveletModel_CShuffle
+    // clang-format off
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     AccData|         CShuffle|     CData|           A|           B|           C|           GEMM| NumGemmK| ABBlockTransfer|       BlockGemm|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|        |        |        |      Type|      Type|        Type|         DataType|      Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+// ######|        |        |        |          |          |            |                 |          |   Operation|   Operation|   Operation|               |    Stage|                |                |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+// ######|        |        |        |          |          |            |                 |          |            |            |            |               |         |                |                |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, AccDataType,              F16, CDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,             256,             256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,                S<1, 32, 1,8>,               8>;
+// clang-format on
+
+using DeviceGemmInstance = DeviceGemmInstance;
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/example/01_gemm/run_gemm_example.inc b/example/01_gemm/run_gemm_example.inc
index 4e2cedb52adc8c7030167ec7f171d91f0796f234..38c72afc60c693ff25fbf8d6679ea7f22ebfdb13 100644
--- a/example/01_gemm/run_gemm_example.inc
+++ b/example/01_gemm/run_gemm_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/02_gemm_bilinear/CMakeLists.txt b/example/02_gemm_bilinear/CMakeLists.txt
index 10ec0f1a71151668e262efcdbaff7100d2d08dfa..7ea4564d1b7e6562c4f34197cc691c0f477610df 100644
--- a/example/02_gemm_bilinear/CMakeLists.txt
+++ b/example/02_gemm_bilinear/CMakeLists.txt
@@ -1 +1,17 @@
-add_example_executable(example_gemm_bilinear_xdl_fp16 gemm_bilinear_xdl_fp16.cpp)
+list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102)
+list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
+    add_example_executable(example_gemm_bilinear_wmma_fp16 gemm_bilinear_wmma_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
+
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+    add_example_executable(example_gemm_bilinear_xdl_fp16 gemm_bilinear_xdl_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp b/example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..877792d7409f48ee562cded6629f7e1bbde650d9
--- /dev/null
+++ b/example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp
@@ -0,0 +1,304 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+struct AlphaBetaAdd
+{
+    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
+
+    template <typename E, typename C, typename D>
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
+        ck::half_t& e, const float& c, const ck::half_t& d) const
+    {
+        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
+    };
+
+    float alpha_;
+    float beta_;
+};
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F16;
+using EDataType        = F16;
+
+using ALayout = Row;
+using BLayout = Col;
+using DLayout = Row;
+using ELayout = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AlphaBetaAdd;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceOpInstance =
+    ck::tensor_operation::device::DeviceGemmMultipleD_Wmma_CShuffle<ALayout,
+                                                                    BLayout,
+                                                                    ck::Tuple<DLayout>,
+                                                                    ELayout,
+                                                                    ADataType,
+                                                                    BDataType,
+                                                                    ck::Tuple<DDataType>,
+                                                                    EDataType,
+                                                                    AccDataType,
+                                                                    CShuffleDataType,
+                                                                    AElementOp,
+                                                                    BElementOp,
+                                                                    CDEElementOp,
+                                                                    GemmSpec,
+                                                                    256,
+                                                                    128,
+                                                                    256,
+                                                                    8,
+                                                                    8,
+                                                                    16,
+                                                                    16,
+                                                                    4,
+                                                                    4,
+                                                                    S<4, 64, 1>,
+                                                                    S<1, 0, 2>,
+                                                                    S<1, 0, 2>,
+                                                                    2,
+                                                                    8,
+                                                                    8,
+                                                                    true,
+                                                                    S<4, 64, 1>,
+                                                                    S<1, 0, 2>,
+                                                                    S<1, 0, 2>,
+                                                                    2,
+                                                                    8,
+                                                                    8,
+                                                                    true,
+                                                                    1,
+                                                                    1,
+                                                                    S<1, 32, 1, 8>,
+                                                                    8>;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = true;
+
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideD = 4096;
+    ck::index_t StrideE = 4096;
+
+    float alpha = 1.0f;
+    float beta  = 1.0f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 6)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        alpha = std::stof(argv[4]);
+        beta  = std::stof(argv[5]);
+    }
+    else if(argc == 13)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        StrideA = std::stoi(argv[7]);
+        StrideB = std::stoi(argv[8]);
+        StrideD = std::stoi(argv[9]);
+        StrideE = std::stoi(argv[10]);
+
+        alpha = std::stof(argv[11]);
+        beta  = std::stof(argv[12]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD, StrideE, alpha, "
+               "beta\n");
+        exit(0);
+    }
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d_device_buf.ToDevice(d_m_n.mData.data());
+    e_device_buf.ToDevice(e_m_n_device_result.mData.data());
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                               b_device_buf.GetDeviceBuffer(),
+                               std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                               e_device_buf.GetDeviceBuffer(),
+                               M,
+                               N,
+                               K,
+                               StrideA,
+                               StrideB,
+                               std::array<ck::index_t, 1>{StrideD},
+                               StrideE,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl;
+
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_m_n({M, N});
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                CShuffleDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+        auto ref_gemm               = ReferenceGemmInstance{};
+        auto ref_invoker            = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp b/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp
index 917b6b1c3142ec8cf7d1c852ea59494fea1842a9..c3e6ef7d5df2e12de69e2f9465f8f3e3709c0859 100644
--- a/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp
+++ b/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/03_gemm_bias_relu/CMakeLists.txt b/example/03_gemm_bias_relu/CMakeLists.txt
index 35c54abac03094f24187df2503aa02b6812c20f3..2f5cba924dfd044a1ff0717d75a8b6e7efc3d16b 100644
--- a/example/03_gemm_bias_relu/CMakeLists.txt
+++ b/example/03_gemm_bias_relu/CMakeLists.txt
@@ -1 +1,8 @@
-add_example_executable(example_gemm_bias_relu_xdl_fp16 gemm_bias_relu_xdl_fp16.cpp)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+    add_example_executable(example_gemm_bias_relu_xdl_fp16 gemm_bias_relu_xdl_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp b/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp
index aee51d05de58b6386a5dd267d45a7e5c12d98276..dffeff23374b6f868fcce952ef8991ca289f59b7 100644
--- a/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp
+++ b/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/04_gemm_add_add_fastgelu/CMakeLists.txt b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
index c75c5ba51e899e4d0cba6564f52a2abb63854dd8..447fa9871b75906aa4b74f15654181f00c129f9b 100644
--- a/example/04_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
@@ -1,17 +1,24 @@
-add_custom_target(example_gemm_add_add_fastgelu_xdl)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+    add_custom_target(example_gemm_add_add_fastgelu_xdl)
 
-add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
-add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
-add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
-endif(USE_BITINT_EXTENSION_INT4)
-add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+    if(USE_BITINT_EXTENSION_INT4)
+       add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
+    endif(USE_BITINT_EXTENSION_INT4)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
 
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
-if(USE_BITINT_EXTENSION_INT4)
-  add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
-endif(USE_BITINT_EXTENSION_INT4)
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
+    if(USE_BITINT_EXTENSION_INT4)
+        add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
+    endif(USE_BITINT_EXTENSION_INT4)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/04_gemm_add_add_fastgelu/common.hpp b/example/04_gemm_add_add_fastgelu/common.hpp
index 3f9375e09266c437b6d5ceb9023a06ba0654aed7..91d17df95fc470ea23011999d0a929b9331d0cc2 100644
--- a/example/04_gemm_add_add_fastgelu/common.hpp
+++ b/example/04_gemm_add_add_fastgelu/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -62,7 +62,7 @@ struct ExecutionConfig final
 };
 
 inline bool
-parse_cmd_args(int argc, char* argv[], ProblemSize& problem_size, ExecutionConfig config)
+parse_cmd_args(int argc, char* argv[], ProblemSize& problem_size, ExecutionConfig& config)
 {
     if(argc == 1)
     {
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp
index 5e50c14dc2b8a8473c3ee5a16a867eebbd147447..e630f6783713517977f63a9a2a1ef20836af2c61 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
@@ -7,10 +7,11 @@ using ADataType        = BF16;
 using BDataType        = BF16;
 using AccDataType      = F32;
 using CShuffleDataType = F32;
-using D0DataType       = BF16;
-using D1DataType       = BF16;
-using DsDataType       = ck::Tuple<D0DataType, D1DataType>;
-using EDataType        = BF16;
+using CDataType  = F32; // C matrix doesn't exsit in GPU memory, this is used for host verification
+using D0DataType = BF16;
+using D1DataType = BF16;
+using DsDataType = ck::Tuple<D0DataType, D1DataType>;
+using EDataType  = BF16;
 
 using ALayout  = Row;
 using BLayout  = Col;
@@ -36,7 +37,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_C
 
 using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
                                                                         BDataType,
-                                                                        AccDataType,
+                                                                        CDataType,
                                                                         AccDataType,
                                                                         AElementOp,
                                                                         BElementOp,
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp
index 6c7ca41444846df8decdaabea32581f83ba7eff5..71f6677bae95d1779c76219f6c8b5bc05c3c1d31 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
@@ -7,10 +7,11 @@ using ADataType        = F16;
 using BDataType        = F16;
 using AccDataType      = F32;
 using CShuffleDataType = F32;
-using D0DataType       = F16;
-using D1DataType       = F16;
-using DsDataType       = ck::Tuple<D0DataType, D1DataType>;
-using EDataType        = F16;
+using CDataType  = F32; // C matrix doesn't exsit in GPU memory, this is used for host verification
+using D0DataType = F16;
+using D1DataType = F16;
+using DsDataType = ck::Tuple<D0DataType, D1DataType>;
+using EDataType  = F16;
 
 using ALayout  = Row;
 using BLayout  = Col;
@@ -36,7 +37,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_C
 
 using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
                                                                         BDataType,
-                                                                        AccDataType,
+                                                                        CDataType,
                                                                         AccDataType,
                                                                         AElementOp,
                                                                         BElementOp,
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp
index 1ef266f23dfb7baebbc9fe7ec988ffcc70cb91b3..4665c3932f419c837125ea3153437302cb9dc022 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp
@@ -1,5 +1,4 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
@@ -7,10 +6,11 @@ using ADataType        = F32;
 using BDataType        = F32;
 using AccDataType      = F32;
 using CShuffleDataType = F32;
-using D0DataType       = F32;
-using D1DataType       = F32;
-using DsDataType       = ck::Tuple<D0DataType, D1DataType>;
-using EDataType        = F32;
+using CDataType  = F32; // C matrix doesn't exsit in GPU memory, this is used for host verification
+using D0DataType = F32;
+using D1DataType = F32;
+using DsDataType = ck::Tuple<D0DataType, D1DataType>;
+using EDataType  = F32;
 
 using ALayout  = Row;
 using BLayout  = Col;
@@ -36,7 +36,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_C
 
 using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
                                                                         BDataType,
-                                                                        AccDataType,
+                                                                        CDataType,
                                                                         AccDataType,
                                                                         AElementOp,
                                                                         BElementOp,
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp
index 8b5bc9879b26181031344cb75d4eab57fbae8a6e..f206bbeb411bb0de0144015d70a892b5587d6f0e 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
@@ -11,10 +11,11 @@ using ADataType        = I4;
 using BDataType        = I4;
 using AccDataType      = I32;
 using CShuffleDataType = I32;
-using D0DataType       = I4;
-using D1DataType       = I4;
-using DsDataType       = ck::Tuple<D0DataType, D1DataType>;
-using EDataType        = I4;
+using CDataType  = I32; // C matrix doesn't exsit in GPU memory, this is used for host verification
+using D0DataType = I4;
+using D1DataType = I4;
+using DsDataType = ck::Tuple<D0DataType, D1DataType>;
+using EDataType  = I4;
 
 using KernelADataType  = I8;
 using KernelBDataType  = I8;
@@ -47,7 +48,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_C
 
 using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
                                                                         BDataType,
-                                                                        AccDataType,
+                                                                        CDataType,
                                                                         AccDataType,
                                                                         AElementOp,
                                                                         BElementOp,
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp
index b236f5e9987e980616951b0b2fd0b59c78ace13d..e46483ab38ac34cb5d4f97602474887d78779327 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
@@ -7,10 +7,11 @@ using ADataType        = I8;
 using BDataType        = I8;
 using AccDataType      = I32;
 using CShuffleDataType = I32;
-using D0DataType       = I8;
-using D1DataType       = I8;
-using DsDataType       = ck::Tuple<D0DataType, D1DataType>;
-using EDataType        = I8;
+using CDataType  = I32; // C matrix doesn't exsit in GPU memory, this is used for host verification
+using D0DataType = I8;
+using D1DataType = I8;
+using DsDataType = ck::Tuple<D0DataType, D1DataType>;
+using EDataType  = I8;
 
 using ALayout  = Row;
 using BLayout  = Col;
@@ -36,7 +37,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_C
 
 using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
                                                                         BDataType,
-                                                                        AccDataType,
+                                                                        CDataType,
                                                                         AccDataType,
                                                                         AElementOp,
                                                                         BElementOp,
diff --git a/example/04_gemm_add_add_fastgelu/run_gemm_add_add_fastgelu_example.inc b/example/04_gemm_add_add_fastgelu/run_gemm_add_add_fastgelu_example.inc
index f3def33b5670cd7269ad0a521695943674c9ff19..cb3147bcd718c115e47163af58e813bd2cf39bfe 100644
--- a/example/04_gemm_add_add_fastgelu/run_gemm_add_add_fastgelu_example.inc
+++ b/example/04_gemm_add_add_fastgelu/run_gemm_add_add_fastgelu_example.inc
@@ -124,7 +124,7 @@ bool run_gemm_add_add_fastgelu(const ProblemSize& problem_size, const ExecutionC
 
     if(config.do_verification)
     {
-        Tensor<AccDataType> c_m_n({M, N});
+        Tensor<CDataType> c_m_n({M, N});
 
         auto ref_gemm    = ReferenceGemmInstance{};
         auto ref_invoker = ref_gemm.MakeInvoker();
diff --git a/example/09_convnd_fwd/CMakeLists.txt b/example/09_convnd_fwd/CMakeLists.txt
index e0a53005b8c2846b6c3baaee40272c9ad0858188..90104c16308204c4c66282c7950c07485df0df1e 100644
--- a/example/09_convnd_fwd/CMakeLists.txt
+++ b/example/09_convnd_fwd/CMakeLists.txt
@@ -1,11 +1,17 @@
-add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
-add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
-add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
-add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
-# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+    add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
+    add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
+    add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
+    add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
+    # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
+    add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
+   set(target 1)
+ endif()
+endforeach()
 
 add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
 add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
 add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
-
diff --git a/example/09_convnd_fwd/convnd_fwd_common.hpp b/example/09_convnd_fwd/convnd_fwd_common.hpp
index 4c594ccdf817a069918b348d7b23c6cc533bde28..109b8f9ee34096601f0dd485db6840e0674c1937 100644
--- a/example/09_convnd_fwd/convnd_fwd_common.hpp
+++ b/example/09_convnd_fwd/convnd_fwd_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_common.hpp b/example/09_convnd_fwd/convnd_fwd_dl_common.hpp
index 855710b9d9a30ec47c74a6f31d1b03ade688e28e..aeddd4fc59fc521a196487bf7c0ea4e88bf6e1c8 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_common.hpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp b/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
index db5a7f0bc3351c9f1670993942a54668c191e56f..564fadcbf8b10acc98e47a67cddd2cb0abe295bc 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
@@ -1,9 +1,9 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp b/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
index 964d784c8592b2e94bc38cc8eab6b7a645d88fc1..0bd90c999233b5392f5914081196ff3b06cc8285 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
@@ -1,9 +1,9 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp b/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
index b0cd88f214c8550c018ad9f4d268370dbd816c7f..9b0c8b31f172dcaf984c7c0c728a65fcb1a0554a 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
@@ -1,9 +1,9 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp
index d55d3154916b0f4ba2fee195b7957904ae3d3139..74cf91d16017bd24096dac6348e5a8815856a51a 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
index d84afba6426b1abe4b10a3b4da1fefe7c2e9272a..f6d69bafd48a4b8bff968c2fb8600793b1310294 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp
index f5acc540cf98b8876caa0d5c0812436dd200deb8..6c3171f6157ac4a83a488af12089a97accfd2c7c 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp
index 8d697976abd43e5d07c49ec4e0819cbdf15ac02b..9977a496d229876c083e0afe99e17ff0fa5fca05 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
index 99f7f2565c748a950841dffa12a1d343df0051ab..bf084b3cc0b6ceab326b2b21e963e0ce2d4bdcad 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc b/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc
index 697ada14ba960f7231db3aa0d5b2482ca38578b8..6474df1c355bc059af5429f96616b2192f46f292 100644
--- a/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc
+++ b/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/09_convnd_fwd/run_convnd_fwd_example.inc b/example/09_convnd_fwd/run_convnd_fwd_example.inc
index 36a68056f1d7a34b4309a06941f0ff1f477eaba8..49852ff6678f7c3fbbfce4ac8b1303876092cf40 100644
--- a/example/09_convnd_fwd/run_convnd_fwd_example.inc
+++ b/example/09_convnd_fwd/run_convnd_fwd_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
index 98941b4db53b022388023e5a0eba10c99b5881be..9577b4569a38b4f4fa5d7d859f0f336920bb3100 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
@@ -1,16 +1,20 @@
-add_custom_target(example_convnd_fwd_reduce_xdl)
-
-add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
-add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
-add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
-add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
-
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
-
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
-  add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
-endif(USE_BITINT_EXTENSION_INT4)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(example_convnd_fwd_reduce_xdl)
+   add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+   add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+   add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+   add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
+      add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
+   endif(USE_BITINT_EXTENSION_INT4)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp
index 00e370f2968df0409155d5407066e863574613ce..137b0d1ff0fb0a7cb7159d52f54b4cd3496629ae 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cassert>
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp
index 6ff29b4b0ff0f11d6c9d3becbd3aaafdccf3020e..4ccacb0bcee93e72eb1fd23203e9404f1c23c478 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp
index 02c19c2b63bf4f9ee663140b414c89f6221cff51..bf495725e8e77a37019d1d858bf3fe3b69e828d1 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp
index 679bb5c0c45a1e2e74fc56d21ee6f7f73cae76ad..5848785673c340149d622fd782671d5fb174edab 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp
index abdbdaf74d5c0bd2b87da17d315e8712cd896d52..bf7127502faac94e6858d37c389f316b7ead35a6 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp
index cf86afa8e94957c01fc92acb5ed2286fcb52466c..3e1694cbe8cae679ef249b6654efc40f8b1fc1ab 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc b/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc
index b3a3891781766acd824bc1974b88f57fdd85b711..cebfeb51d63eac30243d8e3b0468b821f6fc1eb3 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/12_reduce/reduce_blockwise.cpp b/example/12_reduce/reduce_blockwise.cpp
index a7ee9990c1941a73b632f3cc1d32b14a00897cc2..9a736d4cfac9d1062c568e795592711e30d18586 100644
--- a/example/12_reduce/reduce_blockwise.cpp
+++ b/example/12_reduce/reduce_blockwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <initializer_list>
diff --git a/example/12_reduce/reduce_blockwise_impl.hpp b/example/12_reduce/reduce_blockwise_impl.hpp
index 7bafd2d2bbfbc2a086fe4e09d005bde5cd5db5cf..7f8394a7301cc3da8271e1e8c41482d0b06a2859 100644
--- a/example/12_reduce/reduce_blockwise_impl.hpp
+++ b/example/12_reduce/reduce_blockwise_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -9,6 +9,7 @@
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_reduce.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
@@ -16,7 +17,6 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_reduction.hpp"
 
 #include "reduce_example_common.hpp"
 
@@ -236,29 +236,6 @@ int reduce_blockwise_impl(bool do_verification,
         reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
             static_cast<int32_t>(reduce_total_length));
 
-    if(do_verification)
-    {
-        ReductionHost<InOutDataType,
-                      AccDataType,
-                      InOutDataType,
-                      ReduceOperation,
-                      InElementwiseOperation,
-                      AccElementwiseOperation,
-                      Rank,
-                      NumReduceDim,
-                      PropagateNan,
-                      OutputIndex>
-            hostReduce(in.mDesc, out_ref.mDesc, invariantDims, reduceDims);
-
-        hostReduce.Run(alpha,
-                       in.mData.data(),
-                       beta,
-                       out_ref.mData.data(),
-                       out_indices_ref.mData.data(),
-                       in_elementwise_op,
-                       acc_elementwise_op);
-    };
-
     std::array<index_t, Rank> arrInLengths;
     std::array<index_t, Rank> arrInStrides;
     std::array<index_t, NumOutDim> arrOutLengths;
@@ -269,6 +246,48 @@ int reduce_blockwise_impl(bool do_verification,
     ck::ranges::copy(outLengths, arrOutLengths.begin());
     ck::ranges::copy(outStrides, arrOutStrides.begin());
 
+    if(do_verification)
+    {
+        using ReferenceReduceInstance =
+            ck::tensor_operation::host::ReferenceReduce<InOutDataType,
+                                                        AccDataType,
+                                                        InOutDataType,
+                                                        Rank,
+                                                        NumReduceDim,
+                                                        ReduceOperation,
+                                                        InElementwiseOperation,
+                                                        AccElementwiseOperation,
+                                                        PropagateNan,
+                                                        OutputIndex>;
+
+        auto reduce_ref = ReferenceReduceInstance{};
+
+        auto argument_ptr_ref = reduce_ref.MakeArgumentPointer(arrInLengths,
+                                                               arrInStrides,
+                                                               arrOutLengths,
+                                                               arrOutStrides,
+                                                               reduceDims,
+                                                               static_cast<double>(alpha),
+                                                               static_cast<double>(beta),
+                                                               in.mData.data(),
+                                                               nullptr,
+                                                               out_ref.mData.data(),
+                                                               out_indices_ref.mData.data(),
+                                                               in_elementwise_op,
+                                                               acc_elementwise_op);
+
+        if(!reduce_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters not supported by the reduce reference, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = reduce_ref.MakeInvokerPointer();
+
+        invoker_ptr_ref->Run(argument_ptr_ref.get());
+    };
+
     auto reduce = DeviceReduceInstance{};
 
     auto argument_ptr = reduce.MakeArgumentPointer(arrInLengths,
@@ -276,8 +295,8 @@ int reduce_blockwise_impl(bool do_verification,
                                                    arrOutLengths,
                                                    arrOutStrides,
                                                    reduceDims,
-                                                   alpha,
-                                                   beta,
+                                                   static_cast<double>(alpha),
+                                                   static_cast<double>(beta),
                                                    in_dev.GetDeviceBuffer(),
                                                    nullptr,
                                                    out_dev.GetDeviceBuffer(),
@@ -287,9 +306,8 @@ int reduce_blockwise_impl(bool do_verification,
 
     if(!reduce.IsSupportedArgument(argument_ptr.get()))
     {
-        std::cerr
-            << "The runtime parameters seems not supported by the DeviceReduce instance, exiting!"
-            << std::endl;
+        std::cerr << "The runtime parameters not supported by the DeviceReduce instance, exiting!"
+                  << std::endl;
 
         return (-2);
     };
diff --git a/example/12_reduce/reduce_blockwise_two_call.cpp b/example/12_reduce/reduce_blockwise_two_call.cpp
index 39821f240ad5f83c49073a5f188d151fdb2d6473..eb8b5c76d31a1a0fc7a1a7e085416f80be9abf7f 100644
--- a/example/12_reduce/reduce_blockwise_two_call.cpp
+++ b/example/12_reduce/reduce_blockwise_two_call.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -12,13 +12,13 @@
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_reduce.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_reduction.hpp"
 
 using namespace ck;
 using namespace ck::tensor_operation::device;
@@ -97,8 +97,8 @@ int main(int argc, char* argv[])
     // const std::array<int, 3> invariantDims_2 = {0, 1, 2};
 
     // used by the host reduction
-    const std::array<int, 2> reduceDims    = {3, 4};
-    const std::array<int, 3> invariantDims = {0, 1, 2};
+    const std::array<int, 2> reduceDims = {3, 4};
+    // const std::array<int, 3> invariantDims = {0, 1, 2};
 
     const std::vector<size_t> inLengths_1 = {64, 320, 80, 4, 128};
 
@@ -191,29 +191,6 @@ int main(int argc, char* argv[])
         reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
             static_cast<int32_t>(reduce_total_length));
 
-    if(do_verify)
-    {
-        ReductionHost<InOutDataType,
-                      AccDataType,
-                      InOutDataType,
-                      ReduceOperation,
-                      InElementwiseOperation,
-                      AccElementwiseOperation,
-                      5, // Rank
-                      2, // NumReduceDim
-                      PropagateNan,
-                      OutputIndex>
-            hostReduce(in_1.mDesc, out_ref.mDesc, invariantDims, reduceDims);
-
-        hostReduce.Run(alpha,
-                       in_1.mData.data(),
-                       beta,
-                       out_ref.mData.data(),
-                       nullptr,
-                       in_elementwise_op,
-                       acc_elementwise_op);
-    };
-
     std::array<index_t, 5> arrInLengths_1;
     std::array<index_t, 5> arrInStrides_1;
     std::array<index_t, 4> arrInLengths_2;
@@ -228,6 +205,48 @@ int main(int argc, char* argv[])
     ck::ranges::copy(outLengths, arrOutLengths.begin());
     ck::ranges::copy(outStrides, arrOutStrides.begin());
 
+    if(do_verify)
+    {
+        using ReferenceReduceInstance =
+            ck::tensor_operation::host::ReferenceReduce<InOutDataType,
+                                                        AccDataType,
+                                                        InOutDataType,
+                                                        5,
+                                                        2,
+                                                        ReduceOperation,
+                                                        InElementwiseOperation,
+                                                        AccElementwiseOperation,
+                                                        PropagateNan,
+                                                        OutputIndex>;
+
+        auto reduce_ref = ReferenceReduceInstance{};
+
+        auto argument_ptr_ref = reduce_ref.MakeArgumentPointer(arrInLengths_1,
+                                                               arrInStrides_1,
+                                                               arrOutLengths,
+                                                               arrOutStrides,
+                                                               reduceDims,
+                                                               static_cast<double>(alpha),
+                                                               static_cast<double>(beta),
+                                                               in_1.mData.data(),
+                                                               nullptr,
+                                                               out_ref.mData.data(),
+                                                               nullptr,
+                                                               in_elementwise_op,
+                                                               acc_elementwise_op);
+
+        if(!reduce_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters not supported by the reduce reference, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = reduce_ref.MakeInvokerPointer();
+
+        invoker_ptr_ref->Run(argument_ptr_ref.get());
+    };
+
     auto reduce_1 = DeviceReduceInstance_1{};
 
     auto argument_ptr_1 = reduce_1.MakeArgumentPointer(arrInLengths_1,
@@ -235,8 +254,8 @@ int main(int argc, char* argv[])
                                                        arrInLengths_2,
                                                        arrInStrides_2,
                                                        reduceDims_1,
-                                                       1.0f,
-                                                       0.0f,
+                                                       1.0,
+                                                       0.0,
                                                        in_1_dev.GetDeviceBuffer(),
                                                        nullptr,
                                                        in_2_dev.GetDeviceBuffer(),
@@ -246,9 +265,8 @@ int main(int argc, char* argv[])
 
     if(!reduce_1.IsSupportedArgument(argument_ptr_1.get()))
     {
-        std::cout
-            << "The runtime parameters seems not supported by the DeviceReduce instance, exiting!"
-            << std::endl;
+        std::cout << "The runtime parameters seems supported by the DeviceReduce instance, exiting!"
+                  << std::endl;
     };
 
     auto invoker_ptr_1 = reduce_1.MakeInvokerPointer();
@@ -260,8 +278,8 @@ int main(int argc, char* argv[])
                                                        arrOutLengths,
                                                        arrOutStrides,
                                                        reduceDims_2,
-                                                       alpha,
-                                                       beta,
+                                                       static_cast<double>(alpha),
+                                                       static_cast<double>(beta),
                                                        in_2_dev.GetDeviceBuffer(),
                                                        nullptr,
                                                        out_dev.GetDeviceBuffer(),
diff --git a/example/12_reduce/reduce_example_common.hpp b/example/12_reduce/reduce_example_common.hpp
index 05f0a0edb25ba520d475522bd2963bc050ccc010..5f9a48804a7268cef5a4a065d0c875565bdb3e9e 100644
--- a/example/12_reduce/reduce_example_common.hpp
+++ b/example/12_reduce/reduce_example_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/12_reduce/reduce_multiblock_atomic_add.cpp b/example/12_reduce/reduce_multiblock_atomic_add.cpp
index c4d63a3add8b0d8016b11b8834a1385dcca5beb5..120e3f05957fdb9f937b290dfe09458db248f9c0 100644
--- a/example/12_reduce/reduce_multiblock_atomic_add.cpp
+++ b/example/12_reduce/reduce_multiblock_atomic_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <initializer_list>
diff --git a/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp b/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp
index 94867aee41ec07520c81e109e8a5c7c298c0d029..fed62186448d58ce58c2a13c6dd184e205801185 100644
--- a/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp
+++ b/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -9,6 +9,7 @@
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_reduce.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
@@ -16,7 +17,6 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_reduction.hpp"
 
 #include "reduce_example_common.hpp"
 
@@ -149,29 +149,6 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
         reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
             static_cast<int32_t>(reduce_total_length));
 
-    if(do_verification)
-    {
-        ReductionHost<InOutDataType,
-                      AccDataType,
-                      InOutDataType,
-                      ReduceOperation,
-                      InElementwiseOperation,
-                      AccElementwiseOperation,
-                      Rank,
-                      NumReduceDim,
-                      PropagateNan,
-                      false>
-            hostReduce(in.mDesc, out_ref.mDesc, invariantDims, reduceDims);
-
-        hostReduce.Run(alpha,
-                       in.mData.data(),
-                       beta,
-                       out_ref.mData.data(),
-                       nullptr,
-                       in_elementwise_op,
-                       acc_elementwise_op);
-    };
-
     std::array<index_t, Rank> arrInLengths;
     std::array<index_t, Rank> arrInStrides;
     std::array<index_t, NumOutDim> arrOutLengths;
@@ -182,6 +159,48 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
     ck::ranges::copy(outLengths, arrOutLengths.begin());
     ck::ranges::copy(outStrides, arrOutStrides.begin());
 
+    if(do_verification)
+    {
+        using ReferenceReduceInstance =
+            ck::tensor_operation::host::ReferenceReduce<InOutDataType,
+                                                        AccDataType,
+                                                        InOutDataType,
+                                                        Rank,
+                                                        NumReduceDim,
+                                                        ReduceOperation,
+                                                        InElementwiseOperation,
+                                                        AccElementwiseOperation,
+                                                        PropagateNan,
+                                                        false>;
+
+        auto reduce_ref = ReferenceReduceInstance{};
+
+        auto argument_ptr_ref = reduce_ref.MakeArgumentPointer(arrInLengths,
+                                                               arrInStrides,
+                                                               arrOutLengths,
+                                                               arrOutStrides,
+                                                               reduceDims,
+                                                               static_cast<double>(alpha),
+                                                               static_cast<double>(beta),
+                                                               in.mData.data(),
+                                                               nullptr,
+                                                               out_ref.mData.data(),
+                                                               nullptr,
+                                                               in_elementwise_op,
+                                                               acc_elementwise_op);
+
+        if(!reduce_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters not supported by the reduce reference, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = reduce_ref.MakeInvokerPointer();
+
+        invoker_ptr_ref->Run(argument_ptr_ref.get());
+    };
+
     auto reduce = DeviceReduceInstance{};
 
     auto argument_ptr = reduce.MakeArgumentPointer(arrInLengths,
@@ -189,8 +208,8 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
                                                    arrOutLengths,
                                                    arrOutStrides,
                                                    reduceDims,
-                                                   alpha,
-                                                   beta,
+                                                   static_cast<double>(alpha),
+                                                   static_cast<double>(beta),
                                                    in_dev.GetDeviceBuffer(),
                                                    nullptr,
                                                    out_dev.GetDeviceBuffer(),
@@ -200,9 +219,8 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
 
     if(!reduce.IsSupportedArgument(argument_ptr.get()))
     {
-        std::cerr
-            << "The runtime parameters seems not supported by the DeviceReduce instance, exiting!"
-            << std::endl;
+        std::cerr << "The runtime parameters not supported by the DeviceReduce instance, exiting!"
+                  << std::endl;
 
         return (-2);
     };
diff --git a/example/13_pool2d_fwd/pool2d_fwd_common.hpp b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
index b83cb6a96f0b10f0aeff5e57d386f2ac1c3fa4e4..1157ccd38705b92c759a6de1181fe2b135425091 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_common.hpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,115 +17,11 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
 
 template <typename InDataType,
           typename OutDataType,
-          typename AccDataType,
-          typename IndexDataType,
-          ck::ReduceTensorOp ReduceOpId,
-          bool PropagateNan,
-          bool OutputIndex>
-static void pool_host_verify(const Tensor<InDataType>& in,
-                             Tensor<OutDataType>& out,
-                             Tensor<IndexDataType>& out_indices,
-                             const std::array<ck::index_t, 2>& window_spatial_lengths,
-                             const std::array<ck::index_t, 2>& window_strides,
-                             const std::array<ck::index_t, 2>& in_left_pads,
-                             const std::array<ck::index_t, 2>& /*in_right_pads*/)
-{
-    const int32_t reduceLength = window_spatial_lengths[0] * window_spatial_lengths[1];
-
-    using ReduceOperation = typename ck::reduce_binary_operator<ReduceOpId>::opType;
-
-    auto elementwise_ops =
-        ck::reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(reduceLength);
-
-    auto in_elementwise_op  = std::get<0>(elementwise_ops);
-    auto acc_elementwise_op = std::get<1>(elementwise_ops);
-
-    if constexpr(!OutputIndex)
-    {
-        using Accumulation =
-            ck::detail::AccumulateWithNanCheck<PropagateNan, ReduceOperation, AccDataType>;
-
-        auto f_nchw = [&](auto n, auto c, auto ho, auto wo) {
-            auto accuVal = ReduceOperation::template GetIdentityValue<AccDataType>();
-
-            for(ck::index_t y = 0; y < window_spatial_lengths[0]; ++y)
-            {
-                ck::index_t hi = ho * window_strides[0] + y - in_left_pads[0];
-                for(ck::index_t x = 0; x < window_spatial_lengths[1]; ++x)
-                {
-                    ck::index_t wi = wo * window_strides[1] + x - in_left_pads[1];
-                    if(hi >= 0 && hi < static_cast<ck::index_t>(in.mDesc.GetLengths()[2]) &&
-                       wi >= 0 && wi < static_cast<ck::index_t>(in.mDesc.GetLengths()[3]))
-                    {
-                        AccDataType currVal = static_cast<AccDataType>(in(n, c, hi, wi));
-
-                        in_elementwise_op(currVal, currVal);
-
-                        Accumulation::Calculate(accuVal, currVal);
-                    }
-                }
-            }
-
-            acc_elementwise_op(accuVal, accuVal);
-
-            out(n, c, ho, wo) = accuVal;
-        };
-
-        make_ParallelTensorFunctor(f_nchw,
-                                   out.mDesc.GetLengths()[0],
-                                   out.mDesc.GetLengths()[1],
-                                   out.mDesc.GetLengths()[2],
-                                   out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-    }
-    else
-    {
-        using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
-                                                                        ReduceOperation,
-                                                                        AccDataType,
-                                                                        IndexDataType>;
-        auto f_nchw        = [&](auto n, auto c, auto ho, auto wo) {
-            auto accuVal            = ReduceOperation::template GetIdentityValue<AccDataType>();
-            IndexDataType accuIndex = 0;
-
-            for(ck::index_t y = 0; y < window_spatial_lengths[0]; ++y)
-            {
-                ck::index_t hi = ho * window_strides[0] + y - in_left_pads[0];
-                for(ck::index_t x = 0; x < window_spatial_lengths[1]; ++x)
-                {
-                    ck::index_t wi = wo * window_strides[1] + x - in_left_pads[1];
-                    if(hi >= 0 && hi < in.mDesc.GetLengths()[2] && wi >= 0 &&
-                       wi < in.mDesc.GetLengths()[3])
-                    {
-                        AccDataType currVal     = static_cast<AccDataType>(in(n, c, hi, wi));
-                        IndexDataType currIndex = y * window_spatial_lengths[1] + x;
-
-                        in_elementwise_op(currVal, currVal);
-
-                        Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
-                    }
-                }
-            }
-
-            acc_elementwise_op(accuVal, accuVal);
-
-            out(n, c, ho, wo)         = accuVal;
-            out_indices(n, c, ho, wo) = accuIndex;
-        };
-
-        make_ParallelTensorFunctor(f_nchw,
-                                   out.mDesc.GetLengths()[0],
-                                   out.mDesc.GetLengths()[1],
-                                   out.mDesc.GetLengths()[2],
-                                   out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-    };
-}
-
-template <typename InDataType,
-          typename OutDataType,
-          typename AccDataType,
+          typename ComputeDataType,
           typename IndexDataType,
           typename InLayout,
           typename OutLayout,
@@ -150,9 +46,10 @@ bool pool_test(bool do_verification,
 {
     using DevicePoolFwdInstance =
         ck::tensor_operation::device::DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<
-            InDataType,  // InDataType
-            OutDataType, // OutDataType
-            AccDataType, // AccDataType
+            InDataType,      // InDataType
+            OutDataType,     // OutDataType
+            IndexDataType,   // IndexDataType
+            ComputeDataType, // ComputeDataType
             ReduceOpId,
             OutputIndex,
             64, // BlockSize
@@ -165,10 +62,10 @@ bool pool_test(bool do_verification,
     const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
     const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
 
-    const std::array<ck::index_t, 2> window_spatial_lengths{{Y, X}};
-    const std::array<ck::index_t, 2> window_strides{{window_stride_h, window_stride_w}};
-    const std::array<ck::index_t, 2> input_left_pads{{in_left_pad_h, in_left_pad_w}};
-    const std::array<ck::index_t, 2> input_right_pads{{in_right_pad_h, in_right_pad_w}};
+    const std::vector<ck::index_t> window_spatial_lengths{Y, X};
+    const std::vector<ck::index_t> window_strides{window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> input_left_pads{in_left_pad_h, in_left_pad_w};
+    const std::vector<ck::index_t> input_right_pads{in_right_pad_h, in_right_pad_w};
 
     // tensor layout
     auto f_host_tensor_descriptor =
@@ -219,14 +116,16 @@ bool pool_test(bool do_verification,
         static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
         static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
         static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
-        N,
-        C,
-        std::array<ck::index_t, 2>{{Hi, Wi}},
-        std::array<ck::index_t, 2>{{Y, X}},
-        std::array<ck::index_t, 2>{{Ho, Wo}},
+        {N, C, Hi, Wi},
+        {Y, X},
+        {N, C, Ho, Wo},
+        {C * Hi * Wi, 1, Wi * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
         window_strides,
         input_left_pads,
-        input_right_pads);
+        input_right_pads,
+        {2, 3});
 
     if(!pool.IsSupportedArgument(argument_ptr.get()))
     {
@@ -252,19 +151,28 @@ bool pool_test(bool do_verification,
 
     if(do_verification)
     {
-        pool_host_verify<InDataType,
-                         OutDataType,
-                         AccDataType,
-                         IndexDataType,
-                         ReduceOpId,
-                         PropagateNan,
-                         OutputIndex>(in_n_c_hi_wi,
-                                      out_n_c_ho_wo_host,
-                                      out_indices_n_c_ho_wo_host,
-                                      window_spatial_lengths,
-                                      window_strides,
-                                      input_left_pads,
-                                      input_right_pads);
+        using ReferencePoolingFwdInstance =
+            ck::tensor_operation::host::ReferencePoolingFwd<4,
+                                                            2,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ReduceOpId,
+                                                            PropagateNan,
+                                                            OutputIndex>;
+
+        auto ref_pooling          = ReferencePoolingFwdInstance{};
+        auto ref_pooling_invoker  = ref_pooling.MakeInvoker();
+        auto ref_pooling_argument = ref_pooling.MakeArgument(in_n_c_hi_wi,
+                                                             out_n_c_ho_wo_host,
+                                                             out_indices_n_c_ho_wo_host,
+                                                             window_spatial_lengths,
+                                                             window_strides,
+                                                             input_left_pads,
+                                                             input_right_pads);
+
+        ref_pooling_invoker.Run(ref_pooling_argument);
 
         out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
 
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
index 659f3251dcf9f8d7661cf44169458937f0d15088..daf8540d4936806904b8d2630f60fa9c094b1b79 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
@@ -1,8 +1,7 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
-#include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
@@ -10,9 +9,9 @@
 
 #include "pool2d_fwd_common.hpp"
 
-using InDataType  = ck::half_t;
-using OutDataType = ck::half_t;
-using AccDataType = float;
+using InDataType      = ck::half_t;
+using OutDataType     = ck::half_t;
+using ComputeDataType = float;
 
 using IndexDataType = int32_t;
 
@@ -91,7 +90,7 @@ int main(int argc, char* argv[])
 
     bool pass = pool_test<InDataType,
                           OutDataType,
-                          AccDataType,
+                          ComputeDataType,
                           IndexDataType,
                           InLayout,
                           OutLayout,
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
index f47c7ff15147327603b9eeba815e44465313762c..323e3f61f8490e1d1278a025dfb878ea515b2635 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
@@ -1,8 +1,7 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
-#include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/utility/reduction_enums.hpp"
@@ -10,9 +9,9 @@
 
 #include "pool2d_fwd_common.hpp"
 
-using InDataType  = float;
-using OutDataType = float;
-using AccDataType = float;
+using InDataType      = float;
+using OutDataType     = float;
+using ComputeDataType = float;
 
 using IndexDataType = int32_t;
 
@@ -91,7 +90,7 @@ int main(int argc, char* argv[])
 
     bool pass = pool_test<InDataType,
                           OutDataType,
-                          AccDataType,
+                          ComputeDataType,
                           IndexDataType,
                           InLayout,
                           OutLayout,
diff --git a/example/14_gemm_quantization/CMakeLists.txt b/example/14_gemm_quantization/CMakeLists.txt
index ca09c48c10cd404502dcdafc75cc013af90e22ed..72bdff5ab4018c40b6728512d2804f6694e1abe7 100644
--- a/example/14_gemm_quantization/CMakeLists.txt
+++ b/example/14_gemm_quantization/CMakeLists.txt
@@ -1,2 +1,13 @@
-add_example_executable(example_gemm_xdl_bias_relu_quantization_int8 gemm_xdl_bias_relu_quantization_int8.cpp)
-add_example_executable(example_gemm_xdl_quantization_int8 gemm_xdl_quantization_int8.cpp)
\ No newline at end of file
+# dlops
+add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
+
+# xdlops
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_example_executable(example_gemm_xdl_bias_relu_quantization_int8 gemm_xdl_bias_relu_quantization_int8.cpp)
+   add_example_executable(example_gemm_xdl_quantization_int8 gemm_xdl_quantization_int8.cpp)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/14_gemm_quantization/gemm_dl_quantization_int8.cpp b/example/14_gemm_quantization/gemm_dl_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..2585072dfed88f569f983cc3e563c9deef8036be
--- /dev/null
+++ b/example/14_gemm_quantization/gemm_dl_quantization_int8.cpp
@@ -0,0 +1,204 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using I8  = int8_t;
+using I32 = int32_t;
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using ActivationOp = PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
+
+using ADataType        = I8;
+using BDataType        = I8;
+using AccDataType      = I32;
+using CShuffleDataType = I32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = I8;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Dl<
+    ALayout,
+    BLayout,
+    DsLayout,
+    ELayout,
+    ADataType,
+    BDataType,
+    AccDataType,
+    DsDataType,
+    EDataType,
+    AElementOp,
+    BElementOp,
+    CDEElementOp,
+    GemmDefault,
+    256,                 // BlockSize
+    128,                 // MPerBlock
+    128,                 // NPerBlock
+    16,                  // K0PerBlock
+    4,                   // K1
+    4,                   // M1PerThread
+    4,                   // N1PerThread
+    1,                   // KPerThread
+    S<8, 2>,             // M1N1ThreadClusterM1Xs
+    S<8, 2>,             // M1N1ThreadClusterN1Xs
+    S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+    S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+    S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+    S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+    S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+    S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+    S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+    S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+    S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+    S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+    S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+    S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+    S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+    S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+    S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+    5,                   // CThreadTransferSrcDstVectorDim
+    4>;                  // CThreadTransferDstScalarPerVector
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, EDataType, float, PassThrough, PassThrough, CDEElementOp>;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // GEMM shape
+    ck::index_t M = 1024;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
+
+    ck::index_t StrideA = 1024;
+    ck::index_t StrideB = 1024;
+    ck::index_t StrideE = 1024;
+
+    float requant_scale = 0.03;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
+                                            std::vector<std::size_t>({stride, 1_uz}));
+            }
+            else
+            {
+                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
+                                            std::vector<std::size_t>({1_uz, stride}));
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+    b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{requant_scale, ActivationOp{}};
+
+    // do GEMM
+    auto gemm     = DeviceGemmInstance{};
+    auto invoker  = gemm.MakeInvoker();
+    auto argument = gemm.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                      b_device_buf.GetDeviceBuffer(),
+                                      {},
+                                      e_device_buf.GetDeviceBuffer(),
+                                      M,
+                                      N,
+                                      K,
+                                      StrideA,
+                                      StrideB,
+                                      {},
+                                      StrideE,
+                                      a_element_op,
+                                      b_element_op,
+                                      cde_element_op);
+
+    if(!gemm.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << gemm.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument = ref_gemm.MakeArgument(
+            a_m_k, b_k_n, e_m_n_host_result, a_element_op, b_element_op, cde_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp b/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp
index d5f4e6f62c3eb8e58285e81446a5e1332c37c533..aa3e0116954e603ed5cd759820ea49184bb4cb8a 100644
--- a/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp
+++ b/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp b/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp
index 2371737382447d666d55633fbf3ac976c3017cf7..4b207df5c628976c54d8b9c0f53ea140d51c98e5 100644
--- a/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp
+++ b/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/CMakeLists.txt b/example/15_grouped_gemm/CMakeLists.txt
index 67f6160873584e0fe31e6e60f675c695b33c8460..9df256c38d59f5ae48ca9539b5a3d637c9446c8f 100644
--- a/example/15_grouped_gemm/CMakeLists.txt
+++ b/example/15_grouped_gemm/CMakeLists.txt
@@ -4,12 +4,17 @@ add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
 add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
 add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp)
 add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
+add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
+add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
+
 
 add_dependencies(example_grouped_gemm_xdl
                  example_grouped_gemm_xdl_fp32
                  example_grouped_gemm_xdl_fp16
                  example_grouped_gemm_xdl_bfp16
-                 example_grouped_gemm_xdl_int8)
+                 example_grouped_gemm_xdl_int8
+                 example_grouped_gemm_multiple_d_dl_fp16
+                 example_grouped_gemm_xdl_splitk_fp16)
 
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
diff --git a/example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..3e1f7f089371b17fb02ebca53b27b0860c7b9c72
--- /dev/null
+++ b/example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp
@@ -0,0 +1,67 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <algorithm>
+#include <cstddef>
+#include <initializer_list>
+#include <iostream>
+#include <numeric>
+#include <stdexcept>
+#include <string>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/utility/sequence.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType   = F16;
+using BDataType   = F16;
+using AccDataType = F32;
+using DsDataType  = ck::Tuple<>;
+using EDataType   = F16;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::
+        //   ##################| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|     DsData|     EData|           A|           B|          CDE|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        //   ##################|        |        |         |        |      Type|      Type|        Type|       Type|      Type| Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        //   ##################|        |        |         |        |          |          |            |           |          |   Operation|   Operation|    Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        //   ##################|        |        |         |        |          |          |            |           |          |            |            |             |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+  DeviceGroupedGemmMultipleD_Dl< ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>;
+// clang-format on
+
+#include "run_grouped_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_gemm_example(argc, argv); }
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
index 05d572a1f532446bdc0ec43e59b1eeadfe4bd054..680cee1f814db7cef177e87e4d9c03ba745b8e26 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
index 3f78dafa8977b1b57592500138dd8f034b294efd..90a12bc1ddbf27e04a55de23e3fc6845868dac43 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp
index fd93bb5f87d49a96da8973c3f862b8b48bdeb180..28b0fcd0cea6d6aed46ffeabc35c4c5ebc72b864 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp
index faf41bbf0bba8e06fee9982a2f158706401ca45d..60c4a71a35146c31d9b79703faf5d3422942d817 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp
index 7cb09778c521c9dfaed38c39ee5cb7fcff7bbc26..0c96ef56d3c144f19dcb4c7101a8d481d8960ee6 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..743ab96be6291e617cb012957865a6897bd61a91
--- /dev/null
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
@@ -0,0 +1,97 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F16;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemmXdlSplitKCShuffle
+    // clang-format off
+//######| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+//######|        |        |         |        |      Type|      Type|        Type|         DataType|       Type|      Type| Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+//######|        |        |         |        |          |          |            |                 |           |          |   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+//######|        |        |         |        |          |          |            |                 |           |          |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
+// clang-format on
+
+#include "run_grouped_gemm_example.inc"
+
+int main(int argc, char* argv[])
+{
+    ProblemSize problem_size;
+    ExecutionConfig config;
+
+    problem_size.group_count = 16;
+
+    problem_size.Ms = {
+        167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+
+    for(int i = 0; i < problem_size.group_count; i++)
+    {
+        problem_size.Ns.push_back(768);
+        problem_size.Ks.push_back(4608);
+
+        problem_size.stride_As.push_back(problem_size.Ks[i]);
+        problem_size.stride_Bs.push_back(problem_size.Ks[i]);
+        problem_size.stride_Cs.push_back(problem_size.Ns[i]);
+    }
+
+    if(argc == 4)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=n0, 1=yes)\n");
+        exit(0);
+    }
+
+    return !run_grouped_gemm(problem_size, config);
+}
diff --git a/example/15_grouped_gemm/run_grouped_gemm_example.inc b/example/15_grouped_gemm/run_grouped_gemm_example.inc
index 324e1772805ac2f1b9b84872d3b8baec3c39c6b5..320870e0de7cab9dee43cab6542f13ca5341a90b 100644
--- a/example/15_grouped_gemm/run_grouped_gemm_example.inc
+++ b/example/15_grouped_gemm/run_grouped_gemm_example.inc
@@ -147,6 +147,7 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
 #else
         a_tensors_device[i]->ToDevice(a_tensors[i].mData.data());
         b_tensors_device[i]->ToDevice(b_tensors[i].mData.data());
+        c_tensors_device[i]->SetZero();
 #endif
 
         p_a.push_back(a_tensors_device[i]->GetDeviceBuffer());
diff --git a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
index 226656a736637beea41758bf59dd1bd27ed8189f..a42b427c6152e90a0f04b5cc8783927c32dae844 100644
--- a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
+++ b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
@@ -1,40 +1,47 @@
-add_custom_target(example_gemm_reduce_xdl)
-add_custom_target(example_gemm_reduce_xdl_max)
-add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
-add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
-
-add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
-add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
-add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
-add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
-
-add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
-
-add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
-add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
-add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
-add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
-
-add_dependencies(example_gemm_reduce_xdl_max
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(example_gemm_reduce_xdl)
+   add_custom_target(example_gemm_reduce_xdl_max)
+   add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
+   add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
+
+   add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
+   add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
+   add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
+   add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
+
+   add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
+
+   add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
+   add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
+   add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
+   add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
+
+   add_dependencies(example_gemm_reduce_xdl_max
                  example_gemm_max_xdl_bf16
                  example_gemm_max_xdl_fp16
                  example_gemm_max_xdl_fp32
                  example_gemm_max_xdl_int8)
 
-add_dependencies(example_gemm_reduce_xdl_mean_meansquare
+   add_dependencies(example_gemm_reduce_xdl_mean_meansquare
                  example_gemm_mean_meansquare_xdl_fp16
                  example_gemm_mean_meansquare_xdl_fp32
                  example_gemm_mean_meansquare_xdl_bf16
                  example_gemm_add_addsquare_xdl_int8)
 
-add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
+   add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
 
-add_dependencies(example_gemm_reduce_xdl
+   add_dependencies(example_gemm_reduce_xdl
                  example_gemm_reduce_xdl_mean_meansquare
                  example_gemm_reduce_xdl_max
                  example_gemm_add_add_mean_meansquare_xdl)
 
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
-  add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
-endif()
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
+      add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
+   endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp
index eb3832a668ff9829ee76e87e6ea8988b40e6980b..2f6533d4481a2f339417a19f0f290f7fd306ee9c 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp
index e1248002f751abce96072ee64283ca00022a8713..b28e7f85d3137915a2fe438b16aef5447b2d4d24 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp
index c2feffeb8953df030ad2e30cb942bec529b2cac1..b30ce2c48ad32d99dc5c8a4aaf7a3ee9ea774baf 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp
index 363390add3e99c96ef3891bec995cda5f8c3e24b..31e2efd6f635e30e569d97f73b4a5967b4909b82 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp
index de6b7eb480b30d4347018da94ec518ff29fdbeec..d3c7c1d99c06a222bcefae42d94cdbb6fc8e1239 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp
index 9666fc6622cdf3ab7cbab7071ebffd69d18b7668..9a4a6bc6e11a657b9d1eeaa35dd92552e397130d 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp
index 00e0b767a45dd979868b81635a79600d4695c779..1a8457a8bf87cdc82e687246014d2cba73065b8f 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp
index 652c0e6ea6d2bbe3321de2deefb2c186d7252023..5c2706c79ace02e72f2956270c322c4b00782ebf 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp
index 7eee24fed83988698e0e6d46f830f3a9dfdb0257..c119e243702a27df9bba2d6cf3f791f523c98d7d 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp
index c250b996928dd406bde9253ccbf40e6cd9d77347..0f5e588383abf30064d27e452131ceef1a7ad828 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp b/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp
index 62992de59765d3e05d935ac69894c4550a7ef472..1bea1bcf3e0fc1494e07952020db175a1d91bfbf 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <numeric>
 #include <initializer_list>
diff --git a/example/17_convnd_bwd_data/CMakeLists.txt b/example/17_convnd_bwd_data/CMakeLists.txt
index fa4e65d965f16c7049a21dba93e881fa04cd5319..8ab9f37bb360edd0449f31b13d4be958341317dd 100644
--- a/example/17_convnd_bwd_data/CMakeLists.txt
+++ b/example/17_convnd_bwd_data/CMakeLists.txt
@@ -1,5 +1,11 @@
-add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
-target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
-
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
+   target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
+   set(target 1)
+ endif()
+endforeach()
 add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
 target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
diff --git a/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp b/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
index 26fa9e9821fc050c9ab69fe1a170b4c72a98dcd3..4a9d16c5c303e43989c1f32e51c2cbce6f279e5d 100644
--- a/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
+++ b/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -81,32 +81,33 @@ int run_conv_bwd_data(bool do_verification,
     in_device_buf.SetZero();
 
     // do GEMM
-    auto conv     = DeviceConvNdBwdDataInstance{};
-    auto invoker  = conv.MakeInvoker();
-    auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
-                                      static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
-                                      static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-                                      conv_param.N_,
-                                      conv_param.K_,
-                                      conv_param.C_,
-                                      conv_param.input_spatial_lengths_,
-                                      conv_param.filter_spatial_lengths_,
-                                      conv_param.GetOutputSpatialLengths(),
-                                      conv_param.conv_filter_strides_,
-                                      conv_param.conv_filter_dilations_,
-                                      conv_param.input_left_pads_,
-                                      conv_param.input_right_pads_,
-                                      in_element_op,
-                                      wei_element_op,
-                                      out_element_op);
-
-    if(!conv.IsSupportedArgument(argument))
+    auto conv    = DeviceConvNdBwdDataInstance{};
+    auto invoker = conv.MakeInvoker();
+    auto argument =
+        conv.MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+                                 static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
+                                 static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+                                 conv_param.N_,
+                                 conv_param.K_,
+                                 conv_param.C_,
+                                 conv_param.input_spatial_lengths_,
+                                 conv_param.filter_spatial_lengths_,
+                                 conv_param.GetOutputSpatialLengths(),
+                                 conv_param.conv_filter_strides_,
+                                 conv_param.conv_filter_dilations_,
+                                 conv_param.input_left_pads_,
+                                 conv_param.input_right_pads_,
+                                 in_element_op,
+                                 wei_element_op,
+                                 out_element_op);
+
+    if(!conv.IsSupportedArgument(argument.get()))
     {
         std::cout << "Not support,please check parameters or device";
         return 0;
     }
 
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+    float ave_time = invoker.Run(argument.get(), StreamConfig{nullptr, time_kernel});
 
     std::size_t flop      = conv_param.GetFlops();
     std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
diff --git a/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp b/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp
index f0896e977144867421b857afd0a58a239716712b..6b84eaba471a966c89b62d9f3be32d9b0e0834d6 100644
--- a/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp
+++ b/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_bwd_data_common.hpp"
 
diff --git a/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp b/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp
index c4f2c1f02bb9010ab3f5d97166819732e911b68a..c9989c60ac2c74e6821bdf6dd40e703998accf01 100644
--- a/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp
+++ b/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_bwd_data_common.hpp"
 
diff --git a/example/18_batched_gemm_reduce/CMakeLists.txt b/example/18_batched_gemm_reduce/CMakeLists.txt
index 99fc0043d2803494522b446cabf4a3f98ca12a5a..94ed129dc03664043b1a0295f9f1dcfb38a77002 100644
--- a/example/18_batched_gemm_reduce/CMakeLists.txt
+++ b/example/18_batched_gemm_reduce/CMakeLists.txt
@@ -1,2 +1,8 @@
-add_example_executable(example_batched_gemm_reduce_xdl_fp16 batched_gemm_reduce_xdl_fp16.cpp)
-
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_example_executable(example_batched_gemm_reduce_xdl_fp16 batched_gemm_reduce_xdl_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp b/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp
index c2e3602a7bb22e3f31c66bb6045171229584fa14..e363dc5c12dd84a36a67e969b06d29a179679f94 100644
--- a/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp
+++ b/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp b/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
index 9eae27ca6e92996e87f328cb1a32500dc7f1d658..24c8d82f674d97afae8cb1b3dc0274c1c13daf35 100644
--- a/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
+++ b/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
@@ -1,12 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
@@ -23,13 +23,13 @@ using CDataType  = F16;
 using Add = ck::tensor_operation::element_wise::Add;
 
 using DeviceElementwiseAddInstance =
-    ck::tensor_operation::device::DeviceElementwise<ck::Tuple<ABDataType, ABDataType>,
-                                                    ck::Tuple<CDataType>,
-                                                    Add,
-                                                    2,
-                                                    8,
-                                                    ck::Sequence<8, 8>,
-                                                    ck::Sequence<8>>;
+    ck::tensor_operation::device::DeviceElementwiseImpl<ck::Tuple<ABDataType, ABDataType>,
+                                                        ck::Tuple<CDataType>,
+                                                        Add,
+                                                        2,
+                                                        8,
+                                                        ck::Sequence<8, 8>,
+                                                        ck::Sequence<8>>;
 
 template <typename HostTensorA,
           typename HostTensorB,
diff --git a/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp b/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
index 813d38b01ee9c3c44e3eef74bf445d18d87a4f26..3c04c561403d0ecfb0819aa569bdf24511053442 100644
--- a/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
+++ b/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
@@ -1,12 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
@@ -23,13 +23,13 @@ using CDataType  = F16;
 using Add = ck::tensor_operation::element_wise::Add;
 
 using DeviceElementwiseAddInstance =
-    ck::tensor_operation::device::DeviceElementwise<ck::Tuple<ABDataType, ABDataType>,
-                                                    ck::Tuple<CDataType>,
-                                                    Add,
-                                                    3,
-                                                    8,
-                                                    ck::Sequence<1, 8>,
-                                                    ck::Sequence<8>>;
+    ck::tensor_operation::device::DeviceElementwiseImpl<ck::Tuple<ABDataType, ABDataType>,
+                                                        ck::Tuple<CDataType>,
+                                                        Add,
+                                                        3,
+                                                        8,
+                                                        ck::Sequence<1, 8>,
+                                                        ck::Sequence<8>>;
 
 template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor>
 void host_broadcast3D_am_bmnk(HostTensorC& C,
diff --git a/example/19_binary_elementwise/elementwise_add_1d.cpp b/example/19_binary_elementwise/elementwise_add_1d.cpp
index a1ca9378d35edf4e3a89da6af92f48c76a2acede..1ac09641a1e20a6be8bf44f4f1ff0f3be045dedb 100644
--- a/example/19_binary_elementwise/elementwise_add_1d.cpp
+++ b/example/19_binary_elementwise/elementwise_add_1d.cpp
@@ -1,11 +1,11 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
@@ -21,13 +21,13 @@ using CDataType  = F16;
 using Add = ck::tensor_operation::element_wise::Add;
 
 using DeviceElementwiseAddInstance =
-    ck::tensor_operation::device::DeviceElementwise<ck::Tuple<ABDataType, ABDataType>,
-                                                    ck::Tuple<CDataType>,
-                                                    Add,
-                                                    1,
-                                                    8,
-                                                    ck::Sequence<8, 8>,
-                                                    ck::Sequence<8>>;
+    ck::tensor_operation::device::DeviceElementwiseImpl<ck::Tuple<ABDataType, ABDataType>,
+                                                        ck::Tuple<CDataType>,
+                                                        Add,
+                                                        1,
+                                                        8,
+                                                        ck::Sequence<8, 8>,
+                                                        ck::Sequence<8>>;
 
 template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor>
 void host_elementwise1D(
diff --git a/example/19_binary_elementwise/elementwise_add_4d.cpp b/example/19_binary_elementwise/elementwise_add_4d.cpp
index 27e10014815f8adce491cc69b500b48b2fa3957a..e571aa8468008a5f4e7097eee74763a583faf9af 100644
--- a/example/19_binary_elementwise/elementwise_add_4d.cpp
+++ b/example/19_binary_elementwise/elementwise_add_4d.cpp
@@ -1,12 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
@@ -23,13 +23,13 @@ using CDataType  = F16;
 using Add = ck::tensor_operation::element_wise::Add;
 
 using DeviceElementwiseAddInstance =
-    ck::tensor_operation::device::DeviceElementwise<ck::Tuple<ABDataType, ABDataType>,
-                                                    ck::Tuple<CDataType>,
-                                                    Add,
-                                                    4,
-                                                    8,
-                                                    ck::Sequence<8, 8>,
-                                                    ck::Sequence<8>>;
+    ck::tensor_operation::device::DeviceElementwiseImpl<ck::Tuple<ABDataType, ABDataType>,
+                                                        ck::Tuple<CDataType>,
+                                                        Add,
+                                                        4,
+                                                        8,
+                                                        ck::Sequence<8, 8>,
+                                                        ck::Sequence<8>>;
 
 template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor>
 void host_elementwise4D(HostTensorC& C,
diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index 557f7971fa38ee7158c3536a71d325585d47c2d9..db170deccee8d03772bb280b334f33483f6c7a7f 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -1,8 +1,20 @@
-add_custom_target(example_grouped_conv_bwd_weight)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(example_grouped_conv_bwd_weight)
 
-add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
-add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
+   add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
 
-
-add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
+   add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
                                                  example_grouped_conv_bwd_weight_xdl_bf16)
+   set(target 1)
+ endif()
+endforeach()
+
+add_custom_target(example_grouped_conv_bwd_weight_dl)
+
+add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
+
+add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
diff --git a/example/20_grouped_conv_bwd_weight/common.hpp b/example/20_grouped_conv_bwd_weight/common.hpp
index d2a8bed59d61fed543124cf8bbb340e18a94dc01..15727495f0f1f8dfedcb24a0ffa02d3d7aea67e0 100644
--- a/example/20_grouped_conv_bwd_weight/common.hpp
+++ b/example/20_grouped_conv_bwd_weight/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -9,7 +9,6 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..375c309e1c93d30a6d7167d524613c9386510551
--- /dev/null
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp"
+
+using InDataType  = F16;
+using WeiDataType = F16;
+using OutDataType = F16;
+using AccDataType = F32;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = PassThrough;
+
+template <ck::index_t NDimSpatial>
+using DeviceConvBwdWeightInstance =
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl<
+        NDimSpatial,          // NDimSpatial
+        InDataType,           // InDataType
+        WeiDataType,          // WeiDataType
+        OutDataType,          // OutDataType
+        AccDataType,          // AccDataType
+        InElementOp,          // InElementwiseOperation
+        WeiElementOp,         // WeiElementwiseOperation
+        OutElementOp,         // OutElementwiseOperation
+        ConvBwdWeightDefault, // ConvBackwardWeightSpecialization
+        256,                  // BlockSize
+        128,                  // MPerBlock
+        128,                  // NPerBlock
+        16,                   // K0PerBlock
+        2,                    // K1
+        4,                    // M1PerThread
+        4,                    // N1PerThread
+        1,                    // KPerThread
+        S<8, 2>,              // M1N1ThreadClusterM1Xs
+        S<8, 2>,              // M1N1ThreadClusterN1Xs
+        S<1, 8, 1, 1, 2>,     // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<1, 2, 1, 128, 1>,   // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<0, 2, 3, 1, 4>,     // ABlockTransferThreadClusterArrangeOrder
+        S<0, 2, 3, 1, 4>,     // ABlockTransferSrcAccessOrder
+        S<1, 1, 1, 1, 1>,     // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<0, 2, 3, 1, 4>,     // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 1, 1>,     // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<1, 1, 1, 8, 2>,     // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<1, 16, 1, 16, 1>,   // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<0, 1, 4, 2, 3>,     // BBlockTransferThreadClusterArrangeOrder
+        S<0, 1, 4, 2, 3>,     // BBlockTransferSrcAccessOrder
+        S<1, 1, 1, 8, 1>,     // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 4, 2, 3>,     // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 1, 2>,     // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>,  // CThreadTransferSrcDstAccessOrder
+        5,                    // CThreadTransferSrcDstVectorDim
+        4>;                   // CThreadTransferDstScalarPerVector
+
+#include "run_grouped_conv_bwd_weight_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
index 9035309c98ef2c598012c845f07a30bae68ce84f..31e277b5c709a44a3495348c2e9419dafeb7611c 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
@@ -1,8 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
+
 using InDataType = BF16;
 // bf16 kernel use fp32 atomic add to accumulate Weight tensor into global memory
 using WeiDataType = F32;
@@ -13,6 +15,58 @@ using InElementOp  = PassThrough;
 using WeiElementOp = PassThrough;
 using OutElementOp = PassThrough;
 
+template <ck::index_t NDimSpatial>
+using DeviceConvBwdWeightInstance =
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle<
+        NDimSpatial,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWC,
+                                      ck::tensor_layout::convolution::GNHWC,
+                                      ck::tensor_layout::convolution::GNDHWC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GKXC,
+                                      ck::tensor_layout::convolution::GKYXC,
+                                      ck::tensor_layout::convolution::GKZYXC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWK,
+                                      ck::tensor_layout::convolution::GNHWK,
+                                      ck::tensor_layout::convolution::GNDHWK>>,
+        InDataType,           // InDataType
+        WeiDataType,          // WeiDataType
+        OutDataType,          // OutDataType
+        AccDataType,          // AccDataType
+        InElementOp,          // InElementwiseOperation
+        WeiElementOp,         // WeiElementwiseOperation
+        OutElementOp,         // OutElementwiseOperation
+        ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
+        256,                  // BlockSize
+        128,                  // MPerBlock
+        128,                  // NPerBlock
+        4,                    // K0PerBlock
+        8,                    // K1
+        32,                   // MPerXdl
+        32,                   // NPerXdl
+        2,                    // MXdlPerWave
+        2,                    // NXdlPerWave
+        S<1, 4, 16, 4>,       // ABlockTransferThreadClusterLengths_K0_M_K1
+        S<0, 3, 1, 2>,        // ABlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1, 3>,        // ABlockTransferSrcAccessOrder
+        2,                    // ABlockTransferSrcVectorDim
+        8,                    // ABlockTransferSrcScalarPerVector
+        2,                    // ABlockTransferDstScalarPerVector_K1
+        true,                 // ABlockLdsAddExtraM
+        S<1, 4, 16, 4>,       // BBlockTransferThreadClusterLengths_K0_N_K1
+        S<0, 3, 1, 2>,        // BBlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1, 3>,        // BBlockTransferSrcAccessOrder
+        2,                    // BBlockTransferSrcVectorDim
+        8,                    // BBlockTransferSrcScalarPerVector
+        2,                    // BBlockTransferDstScalarPerVector_K1
+        true,                 // BBlockLdsAddExtraN
+        1,                    // CShuffleMXdlPerWavePerShuffle
+        1,                    // CShuffleNXdlPerWavePerShuffle
+        S<1, 32, 1, 4>,       // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
+        128 / (sizeof(WeiDataType) * CHAR_BIT)>; // CBlockTransferScalarPerVector_NWaveNPerXdl
+
 #include "run_grouped_conv_bwd_weight_example.inc"
 
 int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
index 6791b0bf68ed556c3ce93456061bd9d5404fe798..69c831cc54eef84fc0b7ea46f9961573bf3efff4 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
@@ -1,8 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
+
 using InDataType  = F16;
 using WeiDataType = F16;
 using OutDataType = F16;
@@ -12,6 +14,58 @@ using InElementOp  = PassThrough;
 using WeiElementOp = PassThrough;
 using OutElementOp = PassThrough;
 
+template <ck::index_t NDimSpatial>
+using DeviceConvBwdWeightInstance =
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle<
+        NDimSpatial,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWC,
+                                      ck::tensor_layout::convolution::GNHWC,
+                                      ck::tensor_layout::convolution::GNDHWC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GKXC,
+                                      ck::tensor_layout::convolution::GKYXC,
+                                      ck::tensor_layout::convolution::GKZYXC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWK,
+                                      ck::tensor_layout::convolution::GNHWK,
+                                      ck::tensor_layout::convolution::GNDHWK>>,
+        InDataType,           // InDataType
+        WeiDataType,          // WeiDataType
+        OutDataType,          // OutDataType
+        AccDataType,          // AccDataType
+        InElementOp,          // InElementwiseOperation
+        WeiElementOp,         // WeiElementwiseOperation
+        OutElementOp,         // OutElementwiseOperation
+        ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
+        256,                  // BlockSize
+        128,                  // MPerBlock
+        128,                  // NPerBlock
+        4,                    // K0PerBlock
+        8,                    // K1
+        32,                   // MPerXdl
+        32,                   // NPerXdl
+        2,                    // MXdlPerWave
+        2,                    // NXdlPerWave
+        S<1, 4, 16, 4>,       // ABlockTransferThreadClusterLengths_K0_M_K1
+        S<0, 3, 1, 2>,        // ABlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1, 3>,        // ABlockTransferSrcAccessOrder
+        2,                    // ABlockTransferSrcVectorDim
+        8,                    // ABlockTransferSrcScalarPerVector
+        2,                    // ABlockTransferDstScalarPerVector_K1
+        true,                 // ABlockLdsAddExtraM
+        S<1, 4, 16, 4>,       // BBlockTransferThreadClusterLengths_K0_N_K1
+        S<0, 3, 1, 2>,        // BBlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1, 3>,        // BBlockTransferSrcAccessOrder
+        2,                    // BBlockTransferSrcVectorDim
+        8,                    // BBlockTransferSrcScalarPerVector
+        2,                    // BBlockTransferDstScalarPerVector_K1
+        true,                 // BBlockLdsAddExtraN
+        1,                    // CShuffleMXdlPerWavePerShuffle
+        1,                    // CShuffleNXdlPerWavePerShuffle
+        S<1, 32, 1, 4>,       // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
+        128 / (sizeof(WeiDataType) * CHAR_BIT)>; // CBlockTransferScalarPerVector_NWaveNPerXdl
+
 #include "run_grouped_conv_bwd_weight_example.inc"
 
 int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index 5264c856fefbbed2a26b105535063b223926f67a..49bd9fc7f09ab90080b61dd034ce9311bf85f792 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -1,45 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-template <ck::index_t NDimSpatial>
-using DeviceConvBwdWeightInstance =
-    ck::tensor_operation::device::DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<
-        NDimSpatial,          // NDimSpatial
-        InDataType,           // InDataType
-        WeiDataType,          // WeiDataType
-        OutDataType,          // OutDataType
-        AccDataType,          // AccDataType
-        InElementOp,          // InElementwiseOperation
-        WeiElementOp,         // WeiElementwiseOperation
-        OutElementOp,         // OutElementwiseOperation
-        ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
-        256,                  // BlockSize
-        128,                  // MPerBlock
-        128,                  // NPerBlock
-        4,                    // K0PerBlock
-        8,                    // K1
-        32,                   // MPerXdl
-        32,                   // NPerXdl
-        2,                    // MXdlPerWave
-        2,                    // NXdlPerWave
-        S<1, 4, 16, 4>,       // ABlockTransferThreadClusterLengths_K0_M_K1
-        S<0, 3, 1, 2>,        // ABlockTransferThreadClusterArrangeOrder
-        S<0, 2, 1, 3>,        // ABlockTransferSrcAccessOrder
-        2,                    // ABlockTransferSrcVectorDim
-        8,                    // ABlockTransferSrcScalarPerVector
-        2,                    // ABlockTransferDstScalarPerVector_K1
-        true,                 // ABlockLdsAddExtraM
-        S<1, 4, 16, 4>,       // BBlockTransferThreadClusterLengths_K0_N_K1
-        S<0, 3, 1, 2>,        // BBlockTransferThreadClusterArrangeOrder
-        S<0, 2, 1, 3>,        // BBlockTransferSrcAccessOrder
-        2,                    // BBlockTransferSrcVectorDim
-        8,                    // BBlockTransferSrcScalarPerVector
-        2,                    // BBlockTransferDstScalarPerVector_K1
-        true,                 // BBlockLdsAddExtraN
-        1,                    // CShuffleMXdlPerWavePerShuffle
-        1,                    // CShuffleNXdlPerWavePerShuffle
-        S<1, 32, 1, 4>,       // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
-        128 / (sizeof(WeiDataType) * CHAR_BIT)>; // CBlockTransferScalarPerVector_NWaveNPerXdl
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 template <ck::index_t NDimSpatial>
 using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
@@ -54,7 +14,20 @@ template <ck::index_t NDimSpatial>
 bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
                                  const ck::utils::conv::ConvParam& conv_param)
 {
-    constexpr ck::index_t split_k = 2;
+    ck::index_t split_k;
+    // Set split_k = 2 for xdl op, split_k = 1 for dl
+    // Dl op doesn't support split_k > 1
+    // TODO: Add Dl op split_k > 1 support
+    if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+         ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+         ck::get_device_name() == "gfx1102"))
+    {
+        split_k = 2;
+    }
+    else
+    {
+        split_k = 1;
+    }
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<
@@ -102,6 +75,8 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> output_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -112,6 +87,8 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
     range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
     range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+    range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+    range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
     range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
     range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
     range_copy(conv_param.input_left_pads_, begin(input_left_pads));
@@ -130,6 +107,8 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
                                       input_spatial_lengths,
                                       filter_spatial_lengths,
                                       output_spatial_lengths,
+                                      input_strides,
+                                      output_strides,
                                       conv_filter_strides,
                                       conv_filter_dilations,
                                       input_left_pads,
@@ -144,7 +123,7 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
         std::cerr << "wrong! device_conv with the specified compilation parameters does "
                      "not support this Conv problem"
                   << std::endl;
-        return false;
+        return true;
     }
 
     float avg_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
diff --git a/example/21_gemm_layernorm/CMakeLists.txt b/example/21_gemm_layernorm/CMakeLists.txt
index 78d3a5d02a5be1a0c7e6475aa080a5c75047ac9b..ff870e7c6b15bca0c9c91c730020fedd0e61546c 100644
--- a/example/21_gemm_layernorm/CMakeLists.txt
+++ b/example/21_gemm_layernorm/CMakeLists.txt
@@ -1,3 +1,11 @@
-add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_fp16 gemm_bias_relu_add_layernorm_xdl_fp16.cpp)
-add_example_executable(example_gemm_layernorm_xdl_fp16 gemm_layernorm_xdl_fp16.cpp)
-add_example_executable(example_gemm_xdl_layernorm_single_kernel_fp16 gemm_xdl_layernorm_single_kernel_fp16.cpp)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_welford_fp16 gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp)
+   add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_naive_fp16 gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp)
+   add_example_executable(example_gemm_layernorm_xdl_naive_fp16 gemm_layernorm_xdl_naive_fp16.cpp)
+   add_example_executable(example_gemm_xdl_layernorm_naive_single_kernel_fp16 gemm_xdl_layernorm_naive_single_kernel_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_fp16.cpp b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp
similarity index 99%
rename from example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_fp16.cpp
rename to example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp
index e37555e7612f088cac22d6fbb8b9c8805e73d453..96d04dcb37798fb6db3334bd2d802e435fdc93fd 100644
--- a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp
@@ -1,16 +1,15 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
 #include <initializer_list>
-#include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/device_memory.hpp"
@@ -95,7 +94,7 @@ using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataTyp
 using NormalizeFunctor = ck::tensor_operation::element_wise::Normalize;
 
 // A:x, B:E[x], C:E[x^2], D:Gamma, E:Beta , F:y
-using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwise<
+using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwiseImpl<
     ck::Tuple<EDataType,
               R0DataType,
               R1DataType,
@@ -116,7 +115,7 @@ auto f_host_tensor_descriptor2d =
     [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
         using namespace ck::literals;
 
-        if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+        if constexpr(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
         {
             return HostTensorDescriptor({row, col}, {stride, 1_uz});
         }
diff --git a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..fc58ca19f8673aa0b2205df21350f7ddf0b92196
--- /dev/null
+++ b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
@@ -0,0 +1,263 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddReluAdd  = ck::tensor_operation::element_wise::AddReluAdd;
+
+// DataType
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using D0DataType       = F16;
+using D1DataType       = F16;
+using DsDataType       = ck::Tuple<D0DataType, D1DataType>;
+using EMeanVarDataType = F16;
+using GammaDataType    = F16;
+using BetaDataType     = F16;
+using HDataType        = F16;
+
+// Layout
+using ALayout  = Row;
+using BLayout  = Col;
+using D0Layout = Row;
+using D1Layout = Row;
+using DsLayout = ck::Tuple<D0Layout, D1Layout>;
+using HLayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddReluAdd;
+using HElementOp   = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// clang-format off
+using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleDLayernorm_Xdl_CShuffle
+//######| ALayout| BLayout| DsLayout| HLayout|     AData|     BData|     AccData|         CShuffle|     DsData|     EMeanVarData|     GammaData|     BetaData|     HData|           A|           B|          CDE|            H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|           PostShuffle|     PostShuffle|            Layernorm|       Layernorm|
+//######|        |        |         |        |      Type|      Type|        Type|         DataType|       Type|             Type|          Type|         Type|      Type| Elementwise| Elementwise|  Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|  ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|
+//######|        |        |         |        |          |          |            |                 |           |                 |              |             |          |   Operation|   Operation|    Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                  _M_N|            _M_N|                 _M_N|              _M|
+//######|        |        |         |        |          |          |            |                 |           |                 |              |             |          |            |            |             |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                      |                |                     |                |
+        < ALayout, BLayout, DsLayout, HLayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EMeanVarDataType, GammaDataType, BetaDataType, HDataType,  AElementOp,  BElementOp, CDEElementOp,   HElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<32, 8>,               8,             S<8, 32>,               8>;
+// clang-format on
+
+auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
+    return HostTensorDescriptor({len}, {stride});
+};
+
+auto f_host_tensor_descriptor2d =
+    [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+        using namespace ck::literals;
+
+        if constexpr(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+        {
+            return HostTensorDescriptor({row, col}, {stride, 1_uz});
+        }
+        else
+        {
+            return HostTensorDescriptor({row, col}, {1_uz, stride});
+        }
+    };
+
+void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
+                         const Tensor<ADataType>& a_m_k,
+                         const Tensor<BDataType>& b_k_n,
+                         const Tensor<D0DataType>& bias_n,
+                         const Tensor<D1DataType>& d1_m_n,
+                         const Tensor<GammaDataType>& gamma_n,
+                         const Tensor<BetaDataType>& beta_n,
+                         AElementOp a_element_op,
+                         BElementOp b_element_op,
+                         CDEElementOp cde_element_op,
+                         HElementOp h_element_op,
+                         int M,
+                         int N,
+                         AccDataType epsilon = 1e-5)
+{
+    using ReferenceGemm = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                    BDataType,
+                                                                    AccDataType,
+                                                                    AccDataType,
+                                                                    AElementOp,
+                                                                    BElementOp,
+                                                                    PassThrough>;
+
+    using ReferenceLayernorm = ck::tensor_operation::host::ReferenceLayernorm<EMeanVarDataType,
+                                                                              GammaDataType,
+                                                                              BetaDataType,
+                                                                              HDataType,
+                                                                              AccDataType,
+                                                                              HElementOp,
+                                                                              2,
+                                                                              1>;
+
+    Tensor<EMeanVarDataType> e_m_n(HostTensorDescriptor{M, N});
+    Tensor<AccDataType> c_m_n(HostTensorDescriptor{M, N});
+
+    auto ref_gemm         = ReferenceGemm{};
+    auto ref_gemm_invoker = ref_gemm.MakeInvoker();
+
+    auto ref_gemm_argument =
+        ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
+
+    ref_gemm_invoker.Run(ref_gemm_argument);
+
+    for(int n = 0; n < N; ++n)
+    {
+        AccDataType bias = static_cast<AccDataType>(bias_n(n));
+        for(int m = 0; m < M; ++m)
+        {
+            AccDataType e  = static_cast<AccDataType>(e_m_n(m, n));
+            AccDataType d1 = static_cast<AccDataType>(d1_m_n(m, n));
+            cde_element_op(e, c_m_n(m, n), bias, d1);
+            e_m_n(m, n) = static_cast<EMeanVarDataType>(e);
+        }
+    }
+
+    ReferenceLayernorm ref_layernorm;
+    auto ref_layernorm_invoker = ref_layernorm.MakeInvoker();
+
+    auto ref_layernorm_argument = ref_layernorm.MakeArgument(
+        e_m_n, gamma_n, beta_n, h_m_n, h_element_op, {M, N}, {1}, epsilon);
+    ref_layernorm_invoker.Run(ref_layernorm_argument);
+}
+
+int main()
+{
+    bool do_verification = true;
+
+    // GEMM shape
+    ck::index_t M = 1024;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
+
+    ck::index_t StrideA  = K;
+    ck::index_t StrideB  = K;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideD1 = N;
+    ck::index_t StrideH  = N;
+
+    float epsilon = 1e-5;
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor2d(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor2d(K, N, StrideB, BLayout{}));
+    Tensor<D0DataType> d0_n(f_host_tensor_descriptor1d(N, 1));
+    Tensor<D1DataType> d1_m_n(f_host_tensor_descriptor2d(M, N, StrideD1, D1Layout{}));
+    Tensor<GammaDataType> gamma_n(f_host_tensor_descriptor1d(N, 1));
+    Tensor<BetaDataType> beta_n(f_host_tensor_descriptor1d(N, 1));
+    Tensor<HDataType> h_m_n(f_host_tensor_descriptor2d(M, N, StrideH, HLayout{}));
+
+    a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{-1, 1});
+    b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-1, 1});
+    d0_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{-1, 1});
+    d1_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{-1, 1});
+    gamma_n.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{-1, 1});
+    beta_n.GenerateTensorValue(GeneratorTensor_3<BetaDataType>{-1, 1});
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d0_device_buf(sizeof(D0DataType) * d0_n.mDesc.GetElementSpaceSize());
+    DeviceMem d1_device_buf(sizeof(D1DataType) * d1_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem gamma_device_buf(sizeof(GammaDataType) * gamma_n.mDesc.GetElementSpaceSize());
+    DeviceMem beta_device_buf(sizeof(BetaDataType) * beta_n.mDesc.GetElementSpaceSize());
+    DeviceMem h_device_buf(sizeof(HDataType) * h_m_n.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d0_device_buf.ToDevice(d0_n.mData.data());
+    d1_device_buf.ToDevice(d1_m_n.mData.data());
+    gamma_device_buf.ToDevice(gamma_n.mData.data());
+    beta_device_buf.ToDevice(beta_n.mData.data());
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{};
+    auto h_element_op   = HElementOp{};
+
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                               b_device_buf.GetDeviceBuffer(),
+                               {d0_device_buf.GetDeviceBuffer(), d1_device_buf.GetDeviceBuffer()},
+                               gamma_device_buf.GetDeviceBuffer(),
+                               beta_device_buf.GetDeviceBuffer(),
+                               h_device_buf.GetDeviceBuffer(),
+                               M,
+                               N,
+                               K,
+                               StrideA,
+                               StrideB,
+                               {StrideD0, StrideD1},
+                               StrideH,
+                               epsilon,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op,
+                               h_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error("wrong! this device_op instance does not support this problem");
+    }
+
+    size_t workspace_sz = device_op.GetWorkSpaceSize(&argument);
+    DeviceMem workspace_dev(workspace_sz);
+    device_op.SetWorkSpacePointer(&argument, workspace_dev.GetDeviceBuffer());
+
+    invoker.Run(argument, StreamConfig{nullptr, false});
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+        Tensor<HDataType> h_m_n_host(HostTensorDescriptor{M, N});
+        host_gemm_layernorm(h_m_n_host,
+                            a_m_k,
+                            b_k_n,
+                            d0_n,
+                            d1_m_n,
+                            gamma_n,
+                            beta_n,
+                            a_element_op,
+                            b_element_op,
+                            cde_element_op,
+                            h_element_op,
+                            M,
+                            N,
+                            epsilon);
+
+        h_device_buf.FromDevice(h_m_n.mData.data());
+        pass &=
+            ck::utils::check_err(h_m_n, h_m_n_host, "Error: Incorrect results h_m_n", 1e-2, 1e-2);
+    }
+
+    return pass ? 0 : 1;
+}
diff --git a/example/21_gemm_layernorm/gemm_layernorm_xdl_fp16.cpp b/example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp
similarity index 98%
rename from example/21_gemm_layernorm/gemm_layernorm_xdl_fp16.cpp
rename to example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp
index 282c8763eba2de02a4d9b0ebfa3ffc20c629ff4a..bd1d6932aceb72890dcb0f7c35004cc0dccd93af 100644
--- a/example/21_gemm_layernorm/gemm_layernorm_xdl_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp
@@ -1,16 +1,15 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
 #include <initializer_list>
-#include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/device_memory.hpp"
@@ -92,7 +91,7 @@ using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataTyp
 using NormalizeFunctor = ck::tensor_operation::element_wise::Normalize;
 
 // A:x, B:E[x], C:E[x^2], D:Gamma, E:Beta , F:y
-using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwise<
+using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwiseImpl<
     ck::Tuple<EDataType,
               R0DataType,
               R1DataType,
@@ -115,7 +114,7 @@ auto f_host_tensor_descriptor2d =
     [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
         using namespace ck::literals;
 
-        if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+        if constexpr(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
         {
             return HostTensorDescriptor({row, col}, {stride, 1_uz});
         }
diff --git a/example/21_gemm_layernorm/gemm_xdl_layernorm_single_kernel_fp16.cpp b/example/21_gemm_layernorm/gemm_xdl_layernorm_naive_single_kernel_fp16.cpp
similarity index 98%
rename from example/21_gemm_layernorm/gemm_xdl_layernorm_single_kernel_fp16.cpp
rename to example/21_gemm_layernorm/gemm_xdl_layernorm_naive_single_kernel_fp16.cpp
index 3c3e36be6a7ab92b48c578a07c6f4a2462b095a9..90d80f9f034b391f75c498f3a34232edf64f5260 100644
--- a/example/21_gemm_layernorm/gemm_xdl_layernorm_single_kernel_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_xdl_layernorm_naive_single_kernel_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -135,7 +135,7 @@ int main(int argc, char* argv[])
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
             using namespace ck::literals;
 
-            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            if constexpr(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
             {
                 return HostTensorDescriptor({row, col}, {stride, 1_uz});
             }
diff --git a/example/22_cgemm/cgemm_xdl_bf16.cpp b/example/22_cgemm/cgemm_xdl_bf16.cpp
index 92ed90ce4ab3ec09da2e053ceebe4d4ed6bcf36d..fa4482a984f20d203bd8bee68614c8e71ecbf83c 100644
--- a/example/22_cgemm/cgemm_xdl_bf16.cpp
+++ b/example/22_cgemm/cgemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_common.hpp b/example/22_cgemm/cgemm_xdl_common.hpp
index 6aa06b7c32cb476b61a88235e13a12a3d1a15db5..26137a7c2e50d0d5010639592c0a614e2eca607b 100644
--- a/example/22_cgemm/cgemm_xdl_common.hpp
+++ b/example/22_cgemm/cgemm_xdl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <numeric>
 #include <initializer_list>
diff --git a/example/22_cgemm/cgemm_xdl_fp16.cpp b/example/22_cgemm/cgemm_xdl_fp16.cpp
index 11373736ee8b37efb5f4082253d46f02074f011f..89a581e865a56b231711352f9733403ed2945aea 100644
--- a/example/22_cgemm/cgemm_xdl_fp16.cpp
+++ b/example/22_cgemm/cgemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_fp32.cpp b/example/22_cgemm/cgemm_xdl_fp32.cpp
index 0f45c18c4818726f179adda70dc12b5ea7c45b9d..cf9659959990823f10ef0cb30fa0eef928c32adc 100644
--- a/example/22_cgemm/cgemm_xdl_fp32.cpp
+++ b/example/22_cgemm/cgemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_int4.cpp b/example/22_cgemm/cgemm_xdl_int4.cpp
index c26a83baafd375b477845fbdd793c3918d7b7dc5..f69cc2b3cc4f98e94ddbf0b37e8ac7324515a8d6 100644
--- a/example/22_cgemm/cgemm_xdl_int4.cpp
+++ b/example/22_cgemm/cgemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_int8.cpp b/example/22_cgemm/cgemm_xdl_int8.cpp
index 2f24189861d8468848d28b9c0480da7c3f2c4fdd..c4835b853ee75496f963a8e14d7a0190eb4cb1cf 100644
--- a/example/22_cgemm/cgemm_xdl_int8.cpp
+++ b/example/22_cgemm/cgemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/23_softmax/softmax_blockwise.cpp b/example/23_softmax/softmax_blockwise.cpp
index 8854bf047be82b0511edd7d09032fccb99a7e8db..d09e434bcfbb262a83c5167429d41d1ca54391ef 100644
--- a/example/23_softmax/softmax_blockwise.cpp
+++ b/example/23_softmax/softmax_blockwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -56,8 +56,8 @@ class SimpleAppArgs
     int option_index = 0;
 
     public:
-    std::vector<size_t> inLengths   = {8, 128, 2048};
-    std::vector<AccDataType> scales = {2.0f, 2.0f};
+    std::vector<size_t> inLengths = {8, 128, 2048};
+    std::vector<double> scales    = {2.0, 2.0};
 
     bool do_verification = true;
     int init_method      = 2;
@@ -151,8 +151,8 @@ int main(int argc, char* argv[])
     auto inStrides  = in.mDesc.GetStrides();
     auto outStrides = out.mDesc.GetStrides();
 
-    AccDataType alpha = args.scales[0];
-    AccDataType beta  = args.scales[1];
+    double alpha = args.scales[0];
+    double beta  = args.scales[1];
 
     std::cout << "in: " << in.mDesc << std::endl;
     std::cout << "out: " << out.mDesc << std::endl;
@@ -221,8 +221,8 @@ int main(int argc, char* argv[])
     auto argument_ptr = device_instance.MakeArgumentPointer(i_inLengths,
                                                             i_inStrides,
                                                             reduceDims,
-                                                            &alpha,
-                                                            &beta,
+                                                            alpha,
+                                                            beta,
                                                             in_dev.GetDeviceBuffer(),
                                                             out_dev.GetDeviceBuffer(),
                                                             PassThrough{},
diff --git a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
index c934d35019602e634da6f8b6d49c1c3b133c6800..420a7cf74f3186ac62d5dc37346a202178d8d273 100644
--- a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
+++ b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
index 98835f98fa6ecea9c38f1155f5dd03d0787fa0d2..9d606db205dde86d544971b4b7fc4830ba73c568 100644
--- a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
+++ b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/26_contraction/CMakeLists.txt b/example/26_contraction/CMakeLists.txt
index 87f4750e3bf5dd81c001718c388038a5840d8a88..c58751f0dd38e8fcc0df0da428777e7f3c7645b1 100644
--- a/example/26_contraction/CMakeLists.txt
+++ b/example/26_contraction/CMakeLists.txt
@@ -1,2 +1,5 @@
 add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
 add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
+
+add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
+add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
index ea105e4ff2bb9a33830fd1182007d658c52ba7e0..78522160c85a1024ec510197e5ab1069e6f077cf 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -16,6 +16,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -74,141 +75,6 @@ using DeviceOpInstanceMNNN = ck::tensor_operation::device::
 
 using DeviceOpInstance = DeviceOpInstanceKKNN;
 
-// hardcoded for NumDimM == NumDimN == NumDimK == 2
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename EDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
-struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
-{
-    // Argument
-    struct Argument : public ck::tensor_operation::device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_ms_ks,
-                 const Tensor<BDataType>& b_ns_ks,
-                 Tensor<EDataType>& e_ms_ns,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : a_ms_ks_{a_ms_ks},
-              b_ns_ks_{b_ns_ks},
-              e_ms_ns_{e_ms_ns},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_ms_ks_;
-        const Tensor<BDataType>& b_ns_ks_;
-        Tensor<EDataType>& e_ms_ns_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public ck::tensor_operation::device::BaseInvoker
-    {
-        using Argument = ReferenceContraction_M2_N2_K2::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
-                const int K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
-                const int K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
-
-                AccDataType v_acc = 0;
-
-                for(int k0 = 0; k0 < K0; ++k0)
-                {
-                    for(int k1 = 0; k1 < K1; ++k1)
-                    {
-                        AccDataType v_a;
-                        AccDataType v_b;
-
-                        arg.a_element_op_(
-                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
-                        arg.b_element_op_(
-                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
-
-                        v_acc += v_a * v_b;
-                    }
-                }
-
-                AccDataType v_c;
-
-                arg.cde_element_op_(v_c, v_acc);
-
-                arg.e_ms_ns_(m0, m1, n0, n1) = v_c;
-            };
-
-            make_ParallelTensorFunctor(f_ms_ns,
-                                       arg.e_ms_ns_.mDesc.GetLengths()[0],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[1],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[2],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[3])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
-    {
-        return true;
-    }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
-                             const Tensor<BDataType>& b_ns_ks,
-                             Tensor<EDataType>& e_ms_ns,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{a_ms_ks, b_ns_ks, e_ms_ns, a_element_op, b_element_op, cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceContraction_M2_N2_K2"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
@@ -385,22 +251,22 @@ int main(int argc, char* argv[])
     {
         Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
 
-        using ReferenceOpInstance = ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                  NumDimN,
-                                                                  NumDimK,
-                                                                  ADataType,
-                                                                  BDataType,
-                                                                  CShuffleDataType,
-                                                                  AccDataType,
-                                                                  AElementOp,
-                                                                  BElementOp,
-                                                                  PassThrough>;
-
-        auto ref_gemm    = ReferenceOpInstance{};
-        auto ref_invoker = ref_gemm.MakeInvoker();
-
-        auto ref_argument = ref_gemm.MakeArgument(
-            a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op, PassThrough{});
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..6cceed5bc11ed8eb79568a0397bbf7e1b93fd33d
--- /dev/null
+++ b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
@@ -0,0 +1,293 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F64 = double;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F64;
+using BDataType        = F64;
+using AccDataType      = F64;
+using CShuffleDataType = F64;
+using DDataType        = F64;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F64;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// clang-format off
+using DeviceOpInstanceKKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceKNNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceMKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceMNNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+// clang-format on
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float alpha = 1.f;
+    float beta  = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 28)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        d_ms_ns_lengths = {M0, M1, N0, N1};
+        d_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])};
+
+        alpha = std::stof(argv[26]);
+        beta  = std::stof(argv[27]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 7: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
+        printf("arg22 to 25: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg26 to 27: alpha, beta\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+    d_device_buf.ToDevice(d_ms_ns.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop      = std::size_t(2) * M * N * K;
+    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                            sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1),
+                                       d_ms_ns(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/26_contraction/contraction_scale_xdl_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp32.cpp
index 26f176b0591d02a821fa783542e9538791e4cbcd..1574f5d18fb7f5f193cd4adb1c1a88616d7b1437 100644
--- a/example/26_contraction/contraction_scale_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -16,6 +16,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -73,141 +74,6 @@ using DeviceOpInstanceMNN = ck::tensor_operation::device::
 
 using DeviceOpInstance = DeviceOpInstanceKKN;
 
-// hardcoded for NumDimM == NumDimN == NumDimK == 2
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename EDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
-struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
-{
-    // Argument
-    struct Argument : public ck::tensor_operation::device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_ms_ks,
-                 const Tensor<BDataType>& b_ns_ks,
-                 Tensor<EDataType>& e_ms_ns,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : a_ms_ks_{a_ms_ks},
-              b_ns_ks_{b_ns_ks},
-              e_ms_ns_{e_ms_ns},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_ms_ks_;
-        const Tensor<BDataType>& b_ns_ks_;
-        Tensor<EDataType>& e_ms_ns_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public ck::tensor_operation::device::BaseInvoker
-    {
-        using Argument = ReferenceContraction_M2_N2_K2::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
-                const int K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
-                const int K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
-
-                AccDataType v_acc = 0;
-
-                for(int k0 = 0; k0 < K0; ++k0)
-                {
-                    for(int k1 = 0; k1 < K1; ++k1)
-                    {
-                        AccDataType v_a;
-                        AccDataType v_b;
-
-                        arg.a_element_op_(
-                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
-                        arg.b_element_op_(
-                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
-
-                        v_acc += v_a * v_b;
-                    }
-                }
-
-                AccDataType v_c;
-
-                arg.cde_element_op_(v_c, v_acc);
-
-                arg.e_ms_ns_(m0, m1, n0, n1) = v_c;
-            };
-
-            make_ParallelTensorFunctor(f_ms_ns,
-                                       arg.e_ms_ns_.mDesc.GetLengths()[0],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[1],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[2],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[3])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
-    {
-        return true;
-    }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
-                             const Tensor<BDataType>& b_ns_ks,
-                             Tensor<EDataType>& e_ms_ns,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{a_ms_ks, b_ns_ks, e_ms_ns, a_element_op, b_element_op, cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceContraction_M2_N2_K2"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
@@ -368,22 +234,23 @@ int main(int argc, char* argv[])
     {
         Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
 
-        using ReferenceOpInstance = ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                  NumDimN,
-                                                                  NumDimK,
-                                                                  ADataType,
-                                                                  BDataType,
-                                                                  CShuffleDataType,
-                                                                  AccDataType,
-                                                                  AElementOp,
-                                                                  BElementOp,
-                                                                  PassThrough>;
-
-        auto ref_gemm    = ReferenceOpInstance{};
-        auto ref_invoker = ref_gemm.MakeInvoker();
-
-        auto ref_argument = ref_gemm.MakeArgument(
-            a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op, PassThrough{});
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
diff --git a/example/26_contraction/contraction_scale_xdl_fp64.cpp b/example/26_contraction/contraction_scale_xdl_fp64.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..3dacc708877d838dd5fb0113b5d560885b0551ae
--- /dev/null
+++ b/example/26_contraction/contraction_scale_xdl_fp64.cpp
@@ -0,0 +1,276 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F64 = double;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F64;
+using BDataType        = F64;
+using AccDataType      = F64;
+using CShuffleDataType = F64;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F64;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// clang-format off
+using DeviceOpInstanceKKN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceMKN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceMNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+// clang-format on
+
+using DeviceOpInstance = DeviceOpInstanceKKN;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float scale = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 23)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        scale = std::stof(argv[22]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg22: scale\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{scale};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 0>{},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/27_layernorm/CMakeLists.txt b/example/27_layernorm/CMakeLists.txt
index d96deae45e49acdc8b86f9973417d6a77e1e7e31..94c23ce77499e0c110301a3d3ae82b0ec119a7c7 100644
--- a/example/27_layernorm/CMakeLists.txt
+++ b/example/27_layernorm/CMakeLists.txt
@@ -1 +1,2 @@
-add_example_executable(example_layernorm_blockwise layernorm_blockwise.cpp)
+add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
+add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
diff --git a/example/27_layernorm/common.hpp b/example/27_layernorm/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..62a71713df84351fea902cdcf3275786b60f5a23
--- /dev/null
+++ b/example/27_layernorm/common.hpp
@@ -0,0 +1,22 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <getopt.h>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
diff --git a/example/27_layernorm/layernorm_fp16.cpp b/example/27_layernorm/layernorm_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..bb8b954f0acaf2e9d748104a05561f9cb02fc9a0
--- /dev/null
+++ b/example/27_layernorm/layernorm_fp16.cpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
+
+constexpr int Rank         = 2;
+constexpr int NumReduceDim = 1;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
+                                                          GammaDataType,
+                                                          BetaDataType,
+                                                          ComputeDataType,
+                                                          YDataType,
+                                                          PassThrough,
+                                                          Rank,
+                                                          NumReduceDim,
+                                                          256, // BlockSize
+                                                          8,   // ClusterM
+                                                          32,  // ClusterK
+                                                          1,   // SliceM
+                                                          8,   // SliceK
+                                                          1,   // XYVectorDim (0=M, 1=K)
+                                                          8,   // SrcScalarPerVector
+                                                          1,   // GammaVecDim (0=M, 1=K)
+                                                          8,   // GammaScalarPerVector
+                                                          1,   // BetaVecDim (0=M, 1=K)
+                                                          8,   // BetaScalarPerVector
+                                                          8>;  // OutScalarPerVector
+#include "run_layernorm_example.inc"
+
+int main() { return run_groupnorm_example<DeviceInstance>(); }
diff --git a/example/27_layernorm/layernorm_splitk_fp16.cpp b/example/27_layernorm/layernorm_splitk_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e0378d028b343a6a70335cb9163370d1586ebf0a
--- /dev/null
+++ b/example/27_layernorm/layernorm_splitk_fp16.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
+
+constexpr int Rank         = 2;
+constexpr int NumReduceDim = 1;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
+                                                                GammaDataType,
+                                                                BetaDataType,
+                                                                ComputeDataType,
+                                                                YDataType,
+                                                                PassThrough,
+                                                                Rank,
+                                                                NumReduceDim,
+                                                                256, // BlockSize
+                                                                8,   // ClusterM
+                                                                32,  // ClusterK
+                                                                1,   // SliceM
+                                                                8,   // SliceK
+                                                                1,   // XYVectorDim (0=M, 1=K)
+                                                                8,   // XScalarPerVector
+                                                                1,   // GammaVecDim (0=M, 1=K)
+                                                                8,   // GammaScalarPerVector
+                                                                1,   // BetaVecDim (0=M, 1=K)
+                                                                8,   // BetaScalarPerVector
+                                                                8>;  // YScalarPerVector
+
+#include "run_layernorm_example.inc"
+
+int main() { return run_groupnorm_example<DeviceInstance>(); }
diff --git a/example/27_layernorm/layernorm_blockwise.cpp b/example/27_layernorm/run_layernorm_example.inc
similarity index 56%
rename from example/27_layernorm/layernorm_blockwise.cpp
rename to example/27_layernorm/run_layernorm_example.inc
index 147307d9ef0df3c3473791382f242f8c217c2ebe..95200b540aa9f9704c8fad785b352bdf779c4094 100644
--- a/example/27_layernorm/layernorm_blockwise.cpp
+++ b/example/27_layernorm/run_layernorm_example.inc
@@ -1,58 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-#include <getopt.h>
-
-#include "ck/ck.hpp"
-#include "ck/utility/reduction_enums.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
-#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
-
-using XDataType     = ck::half_t;
-using GammaDataType = ck::half_t;
-using BetaDataType  = ck::half_t;
-using YDataType     = ck::half_t;
-using AccDataType   = float;
-using PassThrough   = ck::tensor_operation::element_wise::PassThrough;
-
-constexpr int Rank         = 2;
-constexpr int NumReduceDim = 1;
-
-using DeviceInstance =
-    ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
-                                                          GammaDataType,
-                                                          BetaDataType,
-                                                          AccDataType,
-                                                          YDataType,
-                                                          PassThrough,
-                                                          Rank,
-                                                          NumReduceDim,
-                                                          256, // BlockSize
-                                                          8,   // ClusterM
-                                                          32,  // ClusterK
-                                                          1,   // SliceM
-                                                          8,   // SliceK
-                                                          1,   // SrcVecDim (0=M, 1=K)
-                                                          8,   // SrcScalarPerVector
-                                                          1,   // GammaVecDim (0=M, 1=K)
-                                                          8,   // GammaScalarPerVector
-                                                          1,   // BetaVecDim (0=M, 1=K)
-                                                          8,   // BetaScalarPerVector
-                                                          8>;  // OutScalarPerVector
-
-int main()
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+template <typename DeviceInstance>
+int run_groupnorm_example()
 {
     bool time_kernel = false;
 
@@ -111,6 +63,10 @@ int main()
         return 1;
     };
 
+    size_t workspace_sz = device_instance.GetWorkSpaceSize(argument_ptr.get());
+    DeviceMem workspace_dev(workspace_sz);
+    device_instance.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
     auto invoker_ptr = device_instance.MakeInvokerPointer();
     invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
 
@@ -121,7 +77,7 @@ int main()
                                                                                  GammaDataType,
                                                                                  BetaDataType,
                                                                                  YDataType,
-                                                                                 AccDataType,
+                                                                                 ComputeDataType,
                                                                                  PassThrough,
                                                                                  Rank,
                                                                                  NumReduceDim>;
@@ -133,7 +89,8 @@ int main()
         ref_invoker.Run(ref_argument);
 
         y_dev.FromDevice(y.mData.data());
-        pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results d1", 1e-3, 1e-3);
+        pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3);
     }
+
     return (pass ? 0 : 1);
 }
diff --git a/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp b/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp
index f8e6501eadaee4e5b4156309afebdf16e11f3e8a..24e9b1d9b7d569adf20caf1bd0f0cddafdbd256b 100644
--- a/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp
+++ b/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/29_batched_gemm_bias_e_permute/CMakeLists.txt b/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
index 40470f27d427ac8194bb3cf55430b7fd33438597..32a87dd200fc5f9048fc2eb1f56f343fe5586c64 100644
--- a/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
@@ -1 +1,5 @@
 add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
+
+if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+    add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
+endif()
diff --git a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..62233e535151e6b647e938de04b6225fe487b6c4
--- /dev/null
+++ b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp
@@ -0,0 +1,431 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F16;
+using DDataType        = F16;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F16;
+
+static constexpr ck::index_t NumDimG = 2;
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 1;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Add;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::Default;
+
+static constexpr auto ABSpec = ck::tensor_operation::device::TensorSpecialization::Packed;
+static constexpr auto DESpec = ck::tensor_operation::device::TensorSpecialization::Default;
+
+using DeviceOpInstanceKKNN =
+    ck::tensor_operation::device::DeviceBatchedContractionMultipleD_Wmma_CShuffle<NumDimG,
+                                                                                  NumDimM,
+                                                                                  NumDimN,
+                                                                                  NumDimK,
+                                                                                  ADataType,
+                                                                                  BDataType,
+                                                                                  DsDataType,
+                                                                                  EDataType,
+                                                                                  AccDataType,
+                                                                                  CShuffleDataType,
+                                                                                  AElementOp,
+                                                                                  BElementOp,
+                                                                                  CDEElementOp,
+                                                                                  GemmSpec,
+                                                                                  ABSpec,
+                                                                                  ABSpec,
+                                                                                  DESpec,
+                                                                                  256,
+                                                                                  128,
+                                                                                  256,
+                                                                                  8,
+                                                                                  8,
+                                                                                  16,
+                                                                                  16,
+                                                                                  4,
+                                                                                  4,
+                                                                                  S<4, 64, 1>,
+                                                                                  S<1, 0, 2>,
+                                                                                  S<1, 0, 2>,
+                                                                                  2,
+                                                                                  8,
+                                                                                  8,
+                                                                                  true,
+                                                                                  S<4, 64, 1>,
+                                                                                  S<1, 0, 2>,
+                                                                                  S<1, 0, 2>,
+                                                                                  2,
+                                                                                  8,
+                                                                                  8,
+                                                                                  true,
+                                                                                  1,
+                                                                                  1,
+                                                                                  S<1, 32, 1, 8>,
+                                                                                  8>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+// hardcoded for NumDimM == NumDimN == NumDimK == 2
+template <ck::index_t NumDimG,
+          ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType,
+          typename AccDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          ck::enable_if_t<NumDimG == 2 && NumDimM == 2 && NumDimN == 2 && NumDimK == 1, bool> =
+              false>
+struct ReferenceContraction_G2_M2_N2_K1 : public ck::tensor_operation::device::BaseOperator
+{
+    // Argument
+    struct Argument : public ck::tensor_operation::device::BaseArgument
+    {
+        Argument(const Tensor<ADataType>& a_gs_ms_ks,
+                 const Tensor<BDataType>& b_gs_ns_ks,
+                 Tensor<EDataType>& e_gs_ms_ns,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : a_gs_ms_ks_{a_gs_ms_ks},
+              b_gs_ns_ks_{b_gs_ns_ks},
+              e_gs_ms_ns_{e_gs_ms_ns},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op}
+        {
+        }
+
+        const Tensor<ADataType>& a_gs_ms_ks_;
+        const Tensor<BDataType>& b_gs_ns_ks_;
+        Tensor<EDataType>& e_gs_ms_ns_;
+
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+    };
+
+    // Invoker
+    struct Invoker : public ck::tensor_operation::device::BaseInvoker
+    {
+        using Argument = ReferenceContraction_G2_M2_N2_K1::Argument;
+
+        float Run(const Argument& arg)
+        {
+            auto f_ms_ns = [&](auto g0, auto g1, auto m0, auto m1, auto n0, auto n1) {
+                const int K0 = arg.a_gs_ms_ks_.mDesc.GetLengths()[4];
+
+                AccDataType v_acc = 0;
+
+                for(int k0 = 0; k0 < K0; ++k0)
+                {
+                    AccDataType v_a;
+                    AccDataType v_b;
+
+                    arg.a_element_op_(
+                        v_a,
+                        ck::type_convert<const AccDataType>(arg.a_gs_ms_ks_(g0, g1, m0, m1, k0)));
+                    arg.b_element_op_(
+                        v_b,
+                        ck::type_convert<const AccDataType>(arg.b_gs_ns_ks_(g0, g1, n0, n1, k0)));
+
+                    v_acc += v_a * v_b;
+                }
+
+                AccDataType v_c;
+
+                arg.cde_element_op_(v_c, v_acc);
+
+                arg.e_gs_ms_ns_(g0, g1, m0, m1, n0, n1) = v_c;
+            };
+
+            make_ParallelTensorFunctor(f_ms_ns,
+                                       arg.e_gs_ms_ns_.mDesc.GetLengths()[0],
+                                       arg.e_gs_ms_ns_.mDesc.GetLengths()[1],
+                                       arg.e_gs_ms_ns_.mDesc.GetLengths()[2],
+                                       arg.e_gs_ms_ns_.mDesc.GetLengths()[3],
+                                       arg.e_gs_ms_ns_.mDesc.GetLengths()[4],
+                                       arg.e_gs_ms_ns_.mDesc.GetLengths()[5])(
+                std::thread::hardware_concurrency());
+
+            return 0;
+        }
+
+        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
+    {
+        return true;
+    }
+
+    static auto MakeArgument(const Tensor<ADataType>& a_gs_ms_ks,
+                             const Tensor<BDataType>& b_gs_ns_ks,
+                             Tensor<EDataType>& e_gs_ms_ns,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{
+            a_gs_ms_ks, b_gs_ns_ks, e_gs_ms_ns, a_element_op, b_element_op, cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceContraction_G2_M2_N2_K1"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = true;
+
+    ck::index_t G0 = 1;
+    ck::index_t G1 = 2;
+
+    ck::index_t M0 = 4;
+    ck::index_t M1 = 128;
+
+    ck::index_t N0 = 16;
+    ck::index_t N1 = 256;
+
+    ck::index_t K0 = 2048;
+
+    // A[G0, G1, M0, M1, K0]
+    std::vector<ck::index_t> a_gs_ms_ks_lengths{G0, G1, M0, M1, K0};
+    std::vector<ck::index_t> a_gs_ms_ks_strides{G1 * M0 * M1 * K0, M0 * M1 * K0, M1 * K0, K0, 1};
+    // B[G0, G1, N0, N1, K0]
+    std::vector<ck::index_t> b_gs_ns_ks_lengths{G0, G1, N0, N1, K0};
+    std::vector<ck::index_t> b_gs_ns_ks_strides{G1 * N0 * N1 * K0, N0 * N1 * K0, N1 * K0, K0, 1};
+
+    // D[G0, G1, M0, N0, M1, N1]
+    std::vector<ck::index_t> d_gs_ms_ns_lengths{G0, G1, M0, M1, N0, N1};
+    std::vector<ck::index_t> d_gs_ms_ns_strides{G1 * N0 * N1, N0 * N1, 0, 0, N1, 1};
+    // E[G0, G1, M0, N0, M1, N1]
+    std::vector<ck::index_t> e_gs_ms_ns_lengths{G0, G1, M0, M1, N0, N1};
+    std::vector<ck::index_t> e_gs_ms_ns_strides{
+        G1 * M0 * N0 * M1 * N1, M0 * N0 * M1 * N1, N0 * M1 * N1, N1, M1 * N1, 1};
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        exit(0);
+    }
+    Tensor<ADataType> a_gs_ms_ks(a_gs_ms_ks_lengths, a_gs_ms_ks_strides);
+    Tensor<BDataType> b_gs_ns_ks(b_gs_ns_ks_lengths, b_gs_ns_ks_strides);
+    Tensor<DDataType> d_gs_ms_ns(d_gs_ms_ns_lengths, d_gs_ms_ns_strides);
+    Tensor<EDataType> e_gs_ms_ns_host_result(e_gs_ms_ns_lengths, e_gs_ms_ns_strides);
+    Tensor<EDataType> e_gs_ms_ns_device_result(e_gs_ms_ns_lengths, e_gs_ms_ns_strides);
+    std::cout << "a_gs_ms_ks: " << a_gs_ms_ks.mDesc << std::endl;
+    std::cout << "b_gs_ns_ks: " << b_gs_ns_ks.mDesc << std::endl;
+    std::cout << "d_gs_ms_ns: " << d_gs_ms_ns.mDesc << std::endl;
+    std::cout << "e_gs_ms_ns: " << e_gs_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_gs_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_gs_ms_ns.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
+        break;
+    default:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_gs_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_gs_ms_ns.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
+        break;
+    }
+    DeviceMem a_device_buf(sizeof(ADataType) * a_gs_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_gs_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_gs_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) *
+                           e_gs_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_gs_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_gs_ns_ks.mData.data());
+    d_device_buf.ToDevice(d_gs_ms_ns.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_gs_ms_ks_lengths,
+                                    a_gs_ms_ks_strides,
+                                    b_gs_ns_ks_lengths,
+                                    b_gs_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 1>{d_gs_ms_ns_lengths},
+                                    std::array<std::vector<ck::index_t>, 1>{d_gs_ms_ns_strides},
+                                    e_gs_ms_ns_lengths,
+                                    e_gs_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t G =
+        ck::accumulate_n<ck::index_t>(e_gs_ms_ns_lengths.begin(), NumDimG, 1, std::multiplies<>{});
+
+    ck::index_t M = ck::accumulate_n<ck::index_t>(
+        e_gs_ms_ns_lengths.begin() + NumDimG, NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_gs_ms_ns_lengths.begin() + NumDimG + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_gs_ms_ks_lengths.begin() + NumDimG + NumDimM, NumDimK, 1, std::multiplies<>{});
+    std::cout << "GMNK=" << G << ", " << M << ", " << N << ", " << K << std::endl;
+    std::size_t flop      = std::size_t(2) * G * M * N * K;
+    std::size_t num_btype = sizeof(ADataType) * G * M * K + sizeof(BDataType) * G * K * N +
+                            sizeof(DDataType) * G * M * N + sizeof(EDataType) * G * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_gs_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_gs_ms_ns_lengths, e_gs_ms_ns_strides);
+
+        using ReferenceOpInstance = ReferenceContraction_G2_M2_N2_K1<NumDimG,
+                                                                     NumDimM,
+                                                                     NumDimN,
+                                                                     NumDimK,
+                                                                     ADataType,
+                                                                     BDataType,
+                                                                     CShuffleDataType,
+                                                                     AccDataType,
+                                                                     AElementOp,
+                                                                     BElementOp,
+                                                                     PassThrough>;
+
+        auto ref_gemm    = ReferenceOpInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument = ref_gemm.MakeArgument(
+            a_gs_ms_ks, b_gs_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t g0 = 0; g0 < e_gs_ms_ns_host_result.mDesc.GetLengths()[0]; ++g0)
+        {
+            for(size_t g1 = 0; g1 < e_gs_ms_ns_host_result.mDesc.GetLengths()[1]; ++g1)
+            {
+                for(size_t m0 = 0; m0 < e_gs_ms_ns_host_result.mDesc.GetLengths()[2]; ++m0)
+                {
+                    for(size_t m1 = 0; m1 < e_gs_ms_ns_host_result.mDesc.GetLengths()[3]; ++m1)
+                    {
+                        for(size_t n0 = 0; n0 < e_gs_ms_ns_host_result.mDesc.GetLengths()[4]; ++n0)
+                        {
+                            for(size_t n1 = 0; n1 < e_gs_ms_ns_host_result.mDesc.GetLengths()[5];
+                                ++n1)
+                            {
+                                cde_element_op(e_gs_ms_ns_host_result(g0, g1, m0, m1, n0, n1),
+                                               c_ms_ns_host_result(g0, g1, m0, m1, n0, n1),
+                                               d_gs_ms_ns(g0, g1, m0, m1, n0, n1));
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_gs_ms_ns_device_result, e_gs_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp
index 25d815b9cdfdddc29c699a1bc51dd5191a5246eb..08158bfc250642e032d20fad940bb8853c71d446 100644
--- a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp
+++ b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
index 61b2b2f6f3a8c8fdd866f9f5a80f965293f0dd2e..1c07538b09c5261cfd3e5d34f68bccd03087e6e0 100644
--- a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+++ b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
@@ -1,22 +1,36 @@
-add_custom_target(example_grouped_conv_fwd_multiple_d)
+list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102)
 
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
+   add_custom_target(example_grouped_conv_fwd_multiple_d)
 
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
 
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
 
-  add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
-endif() # USE_BITINT_EXTENSION_INT4
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
+      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
+   endif() # USE_BITINT_EXTENSION_INT4
+   add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
+   set(target 1)
+ endif()
+endforeach()
 
-
-add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
-
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/30_grouped_conv_fwd_multiple_d/common.hpp b/example/30_grouped_conv_fwd_multiple_d/common.hpp
index d6d6dd6ff1caa652461114b7626128456295e962..e60ebee6e4fe2e08db432816b75bbbd4c388946c 100644
--- a/example/30_grouped_conv_fwd_multiple_d/common.hpp
+++ b/example/30_grouped_conv_fwd_multiple_d/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -137,7 +137,7 @@ inline bool parse_cmd_args(int argc,
 
         const ck::index_t num_dim_spatial = std::stoi(argv[4]);
         conv_param                        = ck::utils::conv::parse_conv_param(
-            num_dim_spatial, threshold_to_catch_partial_args, argv);
+            num_dim_spatial, threshold_to_catch_partial_args + 1, argv);
     }
     else
     {
diff --git a/example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp b/example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..ae769ff1d38306f4feed8d3cb00dde79b3f561eb
--- /dev/null
+++ b/example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp
@@ -0,0 +1,355 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <algorithm>
+#include <array>
+#include <iostream>
+#include <string>
+#include <type_traits>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
+
+using BF16 = ck::bhalf_t;
+using FP16 = ck::half_t;
+using FP32 = float;
+#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
+using I4 = ck::int4_t;
+#endif
+using I8  = std::int8_t;
+using I32 = std::int32_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <typename InputLay, typename WeightLay, typename OutputLay>
+struct CommonLayoutSetting
+{
+    using InputLayout  = InputLay;
+    using WeightLayout = WeightLay;
+    using OutputLayout = OutputLay;
+};
+
+template <ck::index_t NDimSpatial>
+struct CommonLayoutSettingSelector;
+
+namespace ctl = ck::tensor_layout::convolution;
+
+template <>
+struct CommonLayoutSettingSelector<1> final
+    : CommonLayoutSetting<ctl::G_NW_C, ctl::G_K_X_C, ctl::G_NW_K>
+{
+};
+
+template <>
+struct CommonLayoutSettingSelector<2> final
+    : CommonLayoutSetting<ctl::G_NHW_C, ctl::G_K_YX_C, ctl::G_NHW_K>
+{
+};
+
+template <>
+struct CommonLayoutSettingSelector<3> final
+    : CommonLayoutSetting<ctl::G_NDHW_C, ctl::G_K_ZYX_C, ctl::G_NDHW_K>
+{
+};
+
+template <ck::index_t NDimSpatial>
+using InputLayout = typename CommonLayoutSettingSelector<NDimSpatial>::InputLayout;
+
+template <ck::index_t NDimSpatial>
+using WeightLayout = typename CommonLayoutSettingSelector<NDimSpatial>::WeightLayout;
+
+template <ck::index_t NDimSpatial>
+using OutputLayout = typename CommonLayoutSettingSelector<NDimSpatial>::OutputLayout;
+
+struct ExecutionConfig final
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = true;
+};
+
+#define DefaultConvParam                                                       \
+    ck::utils::conv::ConvParam                                                 \
+    {                                                                          \
+        2, 32, 2, 256, 192, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, { 1, 1 } \
+    }
+
+inline void print_help_msg()
+{
+    std::cerr << "arg1: verification (0=no, 1=yes)\n"
+              << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
+              << "arg3: time kernel (0=no, 1=yes)\n"
+              << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+}
+
+inline bool parse_cmd_args(int argc,
+                           char* argv[],
+                           ExecutionConfig& config,
+                           ck::utils::conv::ConvParam& conv_param)
+{
+    constexpr int num_execution_config_args =
+        3; // arguments for do_verification, init_method, time_kernel
+    constexpr int num_conv_param_leading_args = 5; // arguments for num_dim_spatial_, G_, N_, K_, C_
+
+    constexpr int threshold_to_catch_partial_args = 1 + num_execution_config_args;
+    constexpr int threshold_to_catch_all_args =
+        threshold_to_catch_partial_args + num_conv_param_leading_args;
+
+    if(argc == 1)
+    {
+        // use default
+    }
+    // catch only ExecutionConfig arguments
+    else if(argc == threshold_to_catch_partial_args)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+    }
+    // catch both ExecutionConfig & ConvParam arguments
+    else if(threshold_to_catch_all_args < argc && ((argc - threshold_to_catch_all_args) % 3 == 0))
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t num_dim_spatial = std::stoi(argv[4]);
+        conv_param                        = ck::utils::conv::parse_conv_param(
+            num_dim_spatial, threshold_to_catch_partial_args + 1, argv);
+    }
+    else
+    {
+        print_help_msg();
+        return false;
+    }
+
+    return true;
+}
+
+inline HostTensorDescriptor make_input_descriptor(const ck::utils::conv::ConvParam& conv_param)
+{
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1:
+        return HostTensorDescriptor(
+            {conv_param.G_, conv_param.N_, conv_param.C_, conv_param.input_spatial_lengths_[0]},
+            {
+                conv_param.C_,                                                        // g
+                conv_param.input_spatial_lengths_[0] * conv_param.G_ * conv_param.C_, // n
+                1,                                                                    // c
+                conv_param.G_ * conv_param.C_                                         // wi
+            });
+
+    case 2:
+        return HostTensorDescriptor(
+            {conv_param.G_,
+             conv_param.N_,
+             conv_param.C_,
+             conv_param.input_spatial_lengths_[0],
+             conv_param.input_spatial_lengths_[1]},
+            {
+                conv_param.C_, // g
+                conv_param.input_spatial_lengths_[0] * conv_param.input_spatial_lengths_[1] *
+                    conv_param.G_ * conv_param.C_,                                    // n
+                1,                                                                    // c
+                conv_param.input_spatial_lengths_[1] * conv_param.G_ * conv_param.C_, // hi
+                conv_param.G_ * conv_param.C_                                         // wi
+            });
+
+    case 3:
+        return HostTensorDescriptor(
+            {conv_param.G_,
+             conv_param.N_,
+             conv_param.C_,
+             conv_param.input_spatial_lengths_[0],
+             conv_param.input_spatial_lengths_[1],
+             conv_param.input_spatial_lengths_[2]},
+            {
+                conv_param.C_, // g
+                conv_param.input_spatial_lengths_[0] * conv_param.input_spatial_lengths_[1] *
+                    conv_param.input_spatial_lengths_[2] * conv_param.G_ * conv_param.C_, // n
+                1,                                                                        // c
+                conv_param.input_spatial_lengths_[1] * conv_param.input_spatial_lengths_[2] *
+                    conv_param.G_ * conv_param.C_,                                    // di
+                conv_param.input_spatial_lengths_[2] * conv_param.G_ * conv_param.C_, // hi
+                conv_param.G_ * conv_param.C_                                         // wi
+            });
+    }
+
+    throw std::runtime_error("unsuppored # dim spatial");
+}
+
+inline HostTensorDescriptor make_weight_descriptor(const ck::utils::conv::ConvParam& conv_param)
+{
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1:
+        return HostTensorDescriptor(
+            {conv_param.G_, conv_param.K_, conv_param.C_, conv_param.filter_spatial_lengths_[0]},
+            {
+                conv_param.K_ * conv_param.filter_spatial_lengths_[0] * conv_param.C_, // g
+                conv_param.filter_spatial_lengths_[0] * conv_param.C_,                 // k
+                1,                                                                     // c
+                conv_param.C_                                                          // x
+            });
+    case 2:
+        return HostTensorDescriptor(
+            {conv_param.G_,
+             conv_param.K_,
+             conv_param.C_,
+             conv_param.filter_spatial_lengths_[0],
+             conv_param.filter_spatial_lengths_[1]},
+            {
+                conv_param.K_ * conv_param.filter_spatial_lengths_[0] *
+                    conv_param.filter_spatial_lengths_[1] * conv_param.C_, // g
+                conv_param.filter_spatial_lengths_[0] * conv_param.filter_spatial_lengths_[1] *
+                    conv_param.C_,                                     // k
+                1,                                                     // c
+                conv_param.filter_spatial_lengths_[1] * conv_param.C_, // y
+                conv_param.C_                                          // x
+            });
+    case 3:
+        return HostTensorDescriptor(
+            {conv_param.G_,
+             conv_param.K_,
+             conv_param.C_,
+             conv_param.filter_spatial_lengths_[0],
+             conv_param.filter_spatial_lengths_[1],
+             conv_param.filter_spatial_lengths_[2]},
+            {
+                conv_param.K_ * conv_param.filter_spatial_lengths_[0] *
+                    conv_param.filter_spatial_lengths_[1] * conv_param.filter_spatial_lengths_[2] *
+                    conv_param.C_, // g
+                conv_param.filter_spatial_lengths_[0] * conv_param.filter_spatial_lengths_[1] *
+                    conv_param.filter_spatial_lengths_[2] * conv_param.C_, // k
+                1,                                                         // c
+                conv_param.filter_spatial_lengths_[1] * conv_param.filter_spatial_lengths_[2] *
+                    conv_param.C_,                                     // z
+                conv_param.filter_spatial_lengths_[2] * conv_param.C_, // y
+                conv_param.C_                                          // x
+            });
+    }
+
+    throw std::runtime_error("unsuppored # dim spatial");
+}
+
+inline HostTensorDescriptor make_bias_descriptor(const ck::utils::conv::ConvParam& conv_param)
+{
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1:
+        return HostTensorDescriptor(
+            {conv_param.G_, conv_param.N_, conv_param.K_, conv_param.output_spatial_lengths_[0]},
+            {
+                conv_param.K_, // g
+                0,             // k
+                1,             // c
+                0              // x
+            });
+    case 2:
+        return HostTensorDescriptor({conv_param.G_,
+                                     conv_param.N_,
+                                     conv_param.K_,
+                                     conv_param.output_spatial_lengths_[0],
+                                     conv_param.output_spatial_lengths_[1]},
+                                    {
+                                        conv_param.K_, // g
+                                        0,             // n
+                                        1,             // k
+                                        0,             // ho
+                                        0              // wo
+                                    });
+    case 3:
+        return HostTensorDescriptor({conv_param.G_,
+                                     conv_param.N_,
+                                     conv_param.K_,
+                                     conv_param.output_spatial_lengths_[0],
+                                     conv_param.output_spatial_lengths_[1],
+                                     conv_param.output_spatial_lengths_[2]},
+                                    {
+                                        conv_param.K_, // g
+                                        0,             // n
+                                        1,             // k
+                                        0,             // z
+                                        0,             // y
+                                        0              // x
+                                    });
+    }
+
+    throw std::runtime_error("unsuppored # dim spatial");
+}
+
+inline HostTensorDescriptor make_output_descriptor(const ck::utils::conv::ConvParam& conv_param)
+{
+
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1:
+        return HostTensorDescriptor(
+            {conv_param.G_, conv_param.N_, conv_param.K_, conv_param.output_spatial_lengths_[0]},
+            {
+                conv_param.K_,                                                         // g
+                conv_param.output_spatial_lengths_[0] * conv_param.G_ * conv_param.K_, // n
+                1,                                                                     // k
+                conv_param.G_ * conv_param.K_                                          // wo
+            });
+    case 2:
+        return HostTensorDescriptor(
+            {conv_param.G_,
+             conv_param.N_,
+             conv_param.K_,
+             conv_param.output_spatial_lengths_[0],
+             conv_param.output_spatial_lengths_[1]},
+            {
+                conv_param.K_, // g
+                conv_param.output_spatial_lengths_[0] * conv_param.output_spatial_lengths_[1] *
+                    conv_param.G_ * conv_param.K_,                                     // n
+                1,                                                                     // k
+                conv_param.output_spatial_lengths_[1] * conv_param.G_ * conv_param.K_, // ho
+                conv_param.G_ * conv_param.K_                                          // wo
+            });
+
+    case 3:
+        return HostTensorDescriptor(
+            {conv_param.G_,
+             conv_param.N_,
+             conv_param.K_,
+             conv_param.output_spatial_lengths_[0],
+             conv_param.output_spatial_lengths_[1],
+             conv_param.output_spatial_lengths_[2]},
+            {
+                conv_param.K_, // g
+                conv_param.output_spatial_lengths_[0] * conv_param.output_spatial_lengths_[1] *
+                    conv_param.output_spatial_lengths_[2] * conv_param.G_ * conv_param.K_, // n
+                1,                                                                         // k
+                conv_param.output_spatial_lengths_[1] * conv_param.output_spatial_lengths_[2] *
+                    conv_param.G_ * conv_param.K_,                                     // do
+                conv_param.output_spatial_lengths_[2] * conv_param.G_ * conv_param.K_, // ho
+                conv_param.G_ * conv_param.K_                                          // wo
+            });
+    }
+
+    throw std::runtime_error("unsuppored # dim spatial");
+}
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..039d25029921491e7e67808e554b8cb3e6eb4745
--- /dev/null
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common_wmma.hpp"
+
+// kernel data types
+using InKernelDataType       = FP16;
+using WeiKernelDataType      = FP16;
+using AccDataType            = FP32;
+using CShuffleDataType       = FP16;
+using BiasKernelDataType     = FP16;
+using ResidualKernelDataType = FP16;
+using OutKernelDataType      = FP16;
+
+// tensor data types
+using InUserDataType  = InKernelDataType;
+using WeiUserDataType = WeiKernelDataType;
+using OutUserDataType = OutKernelDataType;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
+
+#include "run_grouped_conv_fwd_bias_relu_add_wmma_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv); }
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..793324970e700748ac4b18e9fc429a24f70c007d
--- /dev/null
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common_wmma.hpp"
+
+// kernel data types
+using InKernelDataType       = I8;
+using WeiKernelDataType      = I8;
+using AccDataType            = I32;
+using CShuffleDataType       = I8;
+using BiasKernelDataType     = I8;
+using ResidualKernelDataType = I8;
+using OutKernelDataType      = I8;
+
+// tensor data types
+using InUserDataType  = InKernelDataType;
+using WeiUserDataType = WeiKernelDataType;
+using OutUserDataType = OutKernelDataType;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
+
+#include "run_grouped_conv_fwd_bias_relu_add_wmma_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv); }
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
index ee300d073a28b44cf8177f64a7278d5e13ef3130..43c0d57dc2adcae69ecde327137eb204b71fa44d 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp
index 5a9df0b1e880c8f3bc954c07cd5de3639644c4e1..40b4132b358d8259146c2987472dc7e9d6c71091 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
index c2906cc9dd1dbe22c8d420ce0929783fb0be0fa6..e05d384f26b40273a6bfe77222b805289a99e631 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
index 3d5a243e6b9282f9d362c82c0758776341df646f..5494563fdd568d1f51ec9ee9042d094096e8fc8a 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
index eaf680fa438a14ee402e3805c18692ffdcf78c7e..6bf2e8d963c1524bfec0680409580b47e5394bc5 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_xdl_fp16.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_xdl_fp16.cpp
index 6de1daa3d454cf556855322deb4b71198fae5d9b..498eda2442fb737a57d15ec0fb58b71a9049f49e 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_xdl_fp16.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc
index 4561156e0bd3a935c49751802e8c3037d8735228..eb242203eaa65ba7b8fca45c5f1e3e1b9d96c409 100644
--- a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc
+++ b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 template <typename BiasLay, typename ResidualLay>
 struct LayoutSetting
diff --git a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc
new file mode 100644
index 0000000000000000000000000000000000000000..360b2c8947b371b1502afb3af73c469f0d8c3636
--- /dev/null
+++ b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc
@@ -0,0 +1,286 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+template <typename BiasLay, typename ResidualLay>
+struct LayoutSetting
+{
+    using BiasLayout     = BiasLay;
+    using ResidualLayout = ResidualLay;
+};
+
+template <ck::index_t NDimSpatial>
+struct LayoutSettingSelector;
+
+template <>
+struct LayoutSettingSelector<1> final : LayoutSetting<ctl::G_K, ctl::G_NW_K>
+{
+};
+
+template <>
+struct LayoutSettingSelector<2> final : LayoutSetting<ctl::G_K, ctl::G_NHW_K>
+{
+};
+
+template <>
+struct LayoutSettingSelector<3> final : LayoutSetting<ctl::G_K, ctl::G_NDHW_K>
+{
+};
+
+template <ck::index_t NDimSpatial>
+using BiasLayout = typename LayoutSettingSelector<NDimSpatial>::BiasLayout;
+
+template <ck::index_t NDimSpatial>
+using ResidualLayout = typename LayoutSettingSelector<NDimSpatial>::ResidualLayout;
+
+template <ck::index_t NDimSpatial>
+using DeviceConvFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<
+        NDimSpatial,
+        InputLayout<NDimSpatial>,
+        WeightLayout<NDimSpatial>,
+        ck::Tuple<BiasLayout<NDimSpatial>, ResidualLayout<NDimSpatial>>,
+        OutputLayout<NDimSpatial>,
+        InKernelDataType,
+        WeiKernelDataType,
+        ck::Tuple<BiasKernelDataType, ResidualKernelDataType>,
+        OutKernelDataType,
+        AccDataType,
+        CShuffleDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        256,         // BlockSize
+        128,         // MPerBlock
+        128,         // NPerBlock
+        4,           // K0PerBlock
+        8,           // K1
+        16,          // MPerWMMA
+        16,          // NPerWMMA
+        4,           // MRepeat
+        2,           // NRepeat
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        8,           // ABlockTransferSrcScalarPerVector
+        8,           // ABlockTransferDstScalarPerVector_AK1
+        true,        // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        8,           // BBlockTransferSrcScalarPerVector
+        8,           // BBlockTransferDstScalarPerVector_BK1
+        true,        // BBlockLdsExtraN
+        4,
+        2,
+        S<1, 32, 1, 8>,
+        8>;
+
+template <ck::index_t NDimSpatial>
+using HostConvFwdInstance = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
+                                                                         InUserDataType,
+                                                                         WeiUserDataType,
+                                                                         CShuffleDataType,
+                                                                         InElementOp,
+                                                                         WeiElementOp,
+                                                                         PassThrough>;
+
+template <ck::index_t NDimSpatial>
+bool run_grouped_conv_fwd_bias_relu_add(const ExecutionConfig& config,
+                                        const ck::utils::conv::ConvParam& conv_param)
+{
+    static_assert(1 <= NDimSpatial && NDimSpatial <= 3, "Unsupported NDimSpatial");
+
+    const auto in_g_n_c_wis_desc   = make_input_descriptor(conv_param);
+    const auto wei_g_k_c_xs_desc   = make_weight_descriptor(conv_param);
+    const auto bias_g_n_k_wos_desc = make_bias_descriptor(conv_param);
+    const auto out_g_n_k_wos_desc  = make_output_descriptor(conv_param);
+
+    Tensor<InUserDataType> in(in_g_n_c_wis_desc);
+    Tensor<WeiUserDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<OutUserDataType> bias(bias_g_n_k_wos_desc);
+    Tensor<OutUserDataType> residual(bias_g_n_k_wos_desc);
+    Tensor<OutUserDataType> out_host(out_g_n_k_wos_desc);
+    Tensor<OutKernelDataType> out_device(out_g_n_k_wos_desc);
+
+    std::cout << "in: " << in.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "bias: " << bias.mDesc << std::endl;
+    std::cout << "residual: " << residual.mDesc << std::endl;
+    std::cout << "out: " << out_host.mDesc << std::endl;
+
+    switch(config.init_method)
+    {
+    case 0: break;
+    case 1:
+        in.GenerateTensorValue(GeneratorTensor_2<InUserDataType>{-5, 5});
+        wei.GenerateTensorValue(GeneratorTensor_2<WeiUserDataType>{-5, 5});
+        bias.GenerateTensorValue(GeneratorTensor_2<OutUserDataType>{-5, 5});
+        break;
+    default:
+        in.GenerateTensorValue(GeneratorTensor_3<InUserDataType>{0.0, 1.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiUserDataType>{-0.5, 0.5});
+        bias.GenerateTensorValue(GeneratorTensor_3<OutUserDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem in_device_buf(sizeof(InKernelDataType) * in.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiKernelDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem bias_device_buf(sizeof(OutKernelDataType) * bias.mDesc.GetElementSpaceSize());
+    DeviceMem residual_device_buf(sizeof(OutKernelDataType) * residual.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutKernelDataType) * out_device.mDesc.GetElementSpaceSize());
+
+#ifdef BUILD_INT4_EXAMPLE
+    const Tensor<InKernelDataType> in_converted(in);
+    const Tensor<WeiKernelDataType> wei_converted(wei);
+    const Tensor<OutKernelDataType> bias_converted(bias);
+    const Tensor<OutKernelDataType> residual_converted(residual);
+
+    in_device_buf.ToDevice(in_converted.mData.data());
+    wei_device_buf.ToDevice(wei_converted.mData.data());
+    bias_device_buf.ToDevice(bias_converted.mData.data());
+    residual_device_buf.ToDevice(residual_converted.mData.data());
+#else
+    in_device_buf.ToDevice(in.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+    bias_device_buf.ToDevice(bias.mData.data());
+    residual_device_buf.ToDevice(residual.mData.data());
+#endif
+
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> d0_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> d0_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> d1_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> d1_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), b_g_k_c_xs_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), b_g_k_c_xs_strides);
+    copy(bias_g_n_k_wos_desc.GetLengths(), d0_g_n_k_wos_lengths);
+    copy(bias_g_n_k_wos_desc.GetStrides(), d0_g_n_k_wos_strides);
+    copy(bias_g_n_k_wos_desc.GetLengths(), d1_g_n_k_wos_lengths);
+    copy(bias_g_n_k_wos_desc.GetStrides(), d1_g_n_k_wos_strides);
+    copy(out_g_n_k_wos_desc.GetLengths(), e_g_n_k_wos_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), e_g_n_k_wos_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    // do Conv
+    auto conv    = DeviceConvFwdInstance<NDimSpatial>{};
+    auto invoker = conv.MakeInvoker();
+    auto argument =
+        conv.MakeArgument(in_device_buf.GetDeviceBuffer(),
+                          wei_device_buf.GetDeviceBuffer(),
+                          std::array<const void*, 2>{bias_device_buf.GetDeviceBuffer(),
+                                                     residual_device_buf.GetDeviceBuffer()},
+                          out_device_buf.GetDeviceBuffer(),
+                          a_g_n_c_wis_lengths,
+                          a_g_n_c_wis_strides,
+                          b_g_k_c_xs_lengths,
+                          b_g_k_c_xs_strides,
+                          std::array<std::array<ck::index_t, NDimSpatial + 3>, 2>{
+                              {d0_g_n_k_wos_lengths, d1_g_n_k_wos_lengths}},
+                          std::array<std::array<ck::index_t, NDimSpatial + 3>, 2>{
+                              {d0_g_n_k_wos_strides, d1_g_n_k_wos_strides}},
+                          e_g_n_k_wos_lengths,
+                          e_g_n_k_wos_strides,
+                          conv_filter_strides,
+                          conv_filter_dilations,
+                          input_left_pads,
+                          input_right_pads,
+                          InElementOp{},
+                          WeiElementOp{},
+                          OutElementOp{});
+
+    if(!conv.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_conv with the specified compilation parameters does "
+            "not support this Conv problem");
+    }
+
+    float avg_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+
+    std::size_t flop      = conv_param.GetFlops();
+    std::size_t num_btype = conv_param.GetByte<InUserDataType, WeiUserDataType, OutUserDataType>();
+
+    float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+    std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << conv.GetTypeString() << std::endl;
+
+    if(config.do_verification)
+    {
+        Tensor<CShuffleDataType> c_host(out_g_n_k_wos_desc);
+
+        auto ref_conv     = HostConvFwdInstance<NDimSpatial>{};
+        auto ref_invoker  = ref_conv.MakeInvoker();
+        auto ref_argument = ref_conv.MakeArgument(in,
+                                                  wei,
+                                                  c_host,
+                                                  conv_param.conv_filter_strides_,
+                                                  conv_param.conv_filter_dilations_,
+                                                  conv_param.input_left_pads_,
+                                                  conv_param.input_right_pads_,
+                                                  InElementOp{},
+                                                  WeiElementOp{},
+                                                  PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        // TODO: implement elementwise operation for host
+        out_host.ForEach([&](auto&, auto idx) {
+            OutElementOp{}(out_host(idx), c_host(idx), bias(idx), residual(idx));
+        });
+
+        out_device_buf.FromDevice(out_device.mData.data());
+
+#ifdef BUILD_INT4_EXAMPLE
+        const Tensor<OutUserDataType> out_device_converted(out_device);
+
+        return ck::utils::check_err(
+            out_device_converted, out_host, "Error: incorrect results!", 1e-5f, 1e-4f);
+#else
+        return ck::utils::check_err(
+            out_device, out_host, "Error: incorrect results!", 1e-5f, 1e-4f);
+#endif
+    }
+
+    return true;
+}
+
+bool run_grouped_conv_fwd_bias_relu_add_example(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_param = DefaultConvParam;
+
+    if(!parse_cmd_args(argc, argv, config, conv_param))
+    {
+        return false;
+    }
+
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1: return run_grouped_conv_fwd_bias_relu_add<1>(config, conv_param);
+    case 2: return run_grouped_conv_fwd_bias_relu_add<2>(config, conv_param);
+    case 3: return run_grouped_conv_fwd_bias_relu_add<3>(config, conv_param);
+    }
+
+    return false;
+}
diff --git a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc
index d087c31af5def501635ef173769851e111dec3cb..58ed69182e0622751dc3fa2823fcb7cfa5b279a8 100644
--- a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc
+++ b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 template <ck::index_t NDimSpatial>
 using DeviceConvFwdInstance =
diff --git a/example/31_batched_gemm_gemm/CMakeLists.txt b/example/31_batched_gemm_gemm/CMakeLists.txt
index d79248251c335c5bc7d5ee2a71c14ea3ac60ac88..83989bcd972c23eec7b814bd09858bf19bb1c3c9 100644
--- a/example/31_batched_gemm_gemm/CMakeLists.txt
+++ b/example/31_batched_gemm_gemm/CMakeLists.txt
@@ -1,8 +1,19 @@
-add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
-add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
-add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
-add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list2 gfx908 gfx90a)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
+   add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
+   add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
+   add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
 
-if(USE_BITINT_EXTENSION_INT4)
-add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
-endif(USE_BITINT_EXTENSION_INT4)
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
+   endif(USE_BITINT_EXTENSION_INT4)
+   set(target 1)
+ endif()
+endforeach()
+
+if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
+   add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+endif()
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp
index 74e0e07e62a99e0700213e80a066d7e15128e516..7605d9c4f8368e369e4c29167f90620984b5a6c8 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp
index d5fadb8081e86a83784f6fad7e7405a7808b65b9..33ed04fb3068b1bf83eb128825ccad355c6de4f9 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp
index 0dd4e0914f479eefae604ceba8115432ce1d3803..e0eb193ad0484c62ac6fa5695ac43aa17c509227 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp
index 1fd93622a1b14a72d8159517a84b9f19d2959723..d166214c3376cd90afd11796a4cb85ea28421861 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp
index 15d98abab7dfa0f1168fd411eeae481054053000..40f87d1f554c4ef2a73ceb32263c3499f733b40a 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc b/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc
index 7e5f1614bcf6c6133d88f9d8dad63ec4be47c9ac..f329146728dd8df3c38af3d2520ac91d43d12d00 100644
--- a/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc
+++ b/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
index 0eb15653306f0179b1d132ddfcd130af6ba13fd6..1d1566d57561afaa84be2de34b11416b5da0571c 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
index 8f1db577c604418b7b1d546b5de251f5a20054a6..bae88d4b8e6f786219e8d68bd0a34f6036dd0d72 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
index 2ce91a8c6023314d61b354614c81ca74c5f95727..a098ce6675e05308fd9de223e807a0e14504a7f5 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
index 1fd2bf69306f5b35dd4ea2ae484c677c6b3c63fc..ce8caf758842225098baac92876ea6996321ff21 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
index f4a8589052f0f149902553884a58cc4c6c79d030..138db14963809cd49294d1cfc2171b038dd10c60 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
index dda9ee9d47735ec1b19d5eea35a40908eaac8e6a..57949242941d7f750d1b121712f1d5cca6cee4d5 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
index 1c0d71c412f47e67e6c4fda887c33b0a74b3ff3c..97caec6053bb7846dd53f37b26c3945a55812967 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_multihead_attention_train_v1.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_multihead_attention_train_v1.cpp
index de39fb1789854f476429dca06687bb2234ad347b..8cb74594cbbe31297404076d20c56c6e59e653cb 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/grouped_multihead_attention_train_v1.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_multihead_attention_train_v1.cpp
@@ -71,13 +71,13 @@ using Scale       = ck::tensor_operation::element_wise::Scale;
 using QKVElementOp = PassThrough;
 using YElementOp   = PassThrough;
 
-using InputDataType    = BF16;
-using OutputDataType   = F32;
-using GemmDataType     = BF16;
+using InputDataType    = F16;
+using OutputDataType   = F16;
+using GemmDataType     = F16;
 using AccDataType      = F32;
 using ShuffleDataType  = F32;
 using LSEDataType      = F32;
-using ZDataType        = INT32; // U16
+using ZDataType        = U16; // INT32
 using Acc0BiasDataType = ck::Tuple<>;
 using Acc1BiasDataType = ck::Tuple<>;
 
@@ -88,7 +88,7 @@ static constexpr ck::index_t NumDimK = 1;
 static constexpr ck::index_t NumDimO = 1;
 // When OutputDataType == F32,      bwd CShuffleBlockTransferScalarPerVector_NPerBlock = 4
 // When OutputDataType == F16/BF16, bwd CShuffleBlockTransferScalarPerVector_NPerBlock = 8
-static constexpr ck::index_t CShuffleBlockTransferScalarPerVector_NPerBlock = 4;
+static constexpr ck::index_t CShuffleBlockTransferScalarPerVector_NPerBlock = 8;
 
 static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKOPadding;
 #if USING_MASK
@@ -103,7 +103,7 @@ static constexpr auto TensorSpecQ   = ck::tensor_operation::device::TensorSpecia
 static constexpr auto TensorSpecK   = ck::tensor_operation::device::TensorSpecialization::Default;
 static constexpr auto TensorSpecV   = ck::tensor_operation::device::TensorSpecialization::Default;
 static constexpr auto TensorSpecY   = ck::tensor_operation::device::TensorSpecialization::Default;
-static constexpr bool Deterministic = true;
+static constexpr bool Deterministic = false;
 
 // DIM should be a multiple of 8.
 // If      DIM <= 32 , ues prototype1 1st template.
diff --git a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc
index 4e43dbdd8fc5a83486cd052ee3e789e05152861f..27602e2313f7aa197e88e1fabeb39245e2fdf5eb 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc
+++ b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
diff --git a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc
index 0b876af952f65f89e9ae55bea344bfa6f61afd5d..fa76faea84e4551ddf8d0617c132dbe7a6045fb3 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc
+++ b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
diff --git a/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc b/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc
index ef2acf61f55fa7f108e174d4586c13ae3dc7419d..ea1e2734a684b61a363eb93ef0e2ff933f900ea5 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc
+++ b/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
diff --git a/example/33_multiple_reduce/dual_reduce_common.hpp b/example/33_multiple_reduce/dual_reduce_common.hpp
index 376b95ea7b307370a8e3f08f228243eebdf4d40f..cd21790be6548abd0f8097613ee650d4407fc400 100644
--- a/example/33_multiple_reduce/dual_reduce_common.hpp
+++ b/example/33_multiple_reduce/dual_reduce_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
@@ -217,8 +217,8 @@ int mean_meansquare_dual_reduce_test(size_t n,
     size_t invariant_total_length = n;
     size_t reduce_total_length    = h * w * c;
 
-    const AccDataType alpha = ck::type_convert<AccDataType>(1.0f);
-    const AccDataType beta  = ck::type_convert<AccDataType>(0.0f);
+    const double alpha = 1.0f;
+    const double beta  = 0.0f;
 
     std::size_t num_thread = 1;
 
@@ -267,8 +267,8 @@ int mean_meansquare_dual_reduce_test(size_t n,
         i_outLengths,
         {i_outStrides, i_outStrides},
         reduceDims,
-        {&alpha, &alpha},
-        {&beta, &beta},
+        {alpha, alpha},
+        {beta, beta},
         in_dev.GetDeviceBuffer(),
         {mean_dev.GetDeviceBuffer(), meansquare_dev.GetDeviceBuffer()},
         ck::make_tuple(InElementwiseOperation_Mean{}, InElementwiseOperation_Meansquare{}),
diff --git a/example/33_multiple_reduce/dual_reduce_multiblock.cpp b/example/33_multiple_reduce/dual_reduce_multiblock.cpp
index 9360599ed9e8a7517bc244e2942d6949c1f71e61..198931749b19b8e8d91bf16c1a733801cb27aeba 100644
--- a/example/33_multiple_reduce/dual_reduce_multiblock.cpp
+++ b/example/33_multiple_reduce/dual_reduce_multiblock.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/33_multiple_reduce/dual_reduce_threadwise.cpp b/example/33_multiple_reduce/dual_reduce_threadwise.cpp
index 56255839e567fa0386ba09b80738f5d0d2abcc23..7609edad3527e8219c2ef6351381d6e53455a67b 100644
--- a/example/33_multiple_reduce/dual_reduce_threadwise.cpp
+++ b/example/33_multiple_reduce/dual_reduce_threadwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/34_batchnorm/CMakeLists.txt b/example/34_batchnorm/CMakeLists.txt
index d964f40d877ddb5dc6e03cf8d28897e244514d96..60824c5f4d7d742941721085be6fc22d638dffe9 100644
--- a/example/34_batchnorm/CMakeLists.txt
+++ b/example/34_batchnorm/CMakeLists.txt
@@ -1,3 +1,4 @@
 add_example_executable(example_batchnorm_forward_training batchnorm_forward_training_nhwc.cpp)
+add_example_executable(example_batchnorm_forward_training_obsolete batchnorm_forward_training_nhwc_obsolete.cpp)
 add_example_executable(example_batchnorm_forward_inferring batchnorm_forward_inferring_nhwc.cpp)
 add_example_executable(example_batchnorm_backward batchnorm_backward_nhwc.cpp)
diff --git a/example/34_batchnorm/batchnorm_backward_nhwc.cpp b/example/34_batchnorm/batchnorm_backward_nhwc.cpp
index a6ca9d150bd918966bae06a63ee4eb9da6be5f3f..3756310fd7dd47b111b9fde743cac844ae2c7669 100644
--- a/example/34_batchnorm/batchnorm_backward_nhwc.cpp
+++ b/example/34_batchnorm/batchnorm_backward_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <limits>
 #include <iostream>
diff --git a/example/34_batchnorm/batchnorm_common.hpp b/example/34_batchnorm/batchnorm_common.hpp
index bdbc8ea8b88f40de2f25a2ec8c5a74ab5e38fd74..a1b8d253bf061c02511945b98eac9343c0d1a236 100644
--- a/example/34_batchnorm/batchnorm_common.hpp
+++ b/example/34_batchnorm/batchnorm_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp b/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp
index dc2984851a02e0020c8c4cfa6d59e61bfe49c593..6a8002025a60b277c587190a95855cc262c33e17 100644
--- a/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp
+++ b/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <limits>
 #include <iostream>
diff --git a/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp b/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
index da36d65a2954ee1ac3a94c617a219e6a7c2baf44..b27358fd9de4547dc8fa8532981f097713eb1e07 100644
--- a/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
+++ b/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <limits>
 #include <iostream>
@@ -414,7 +414,7 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
         (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
 
         y_dev.FromDevice(y.mData.data());
-        pass = pass && ck::utils::check_err(y, y_ref);
+        pass = pass && ck::utils::check_err(y, y_ref, "Incorrect normalized output values");
 
         if(updateMovingAverage)
         {
@@ -424,8 +424,12 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
             resultRunningMean_dev.FromDevice(resultRunningMean.mData.data());
             resultRunningVariance_dev.FromDevice(resultRunningVariance.mData.data());
 
-            pass = pass && ck::utils::check_err(resultRunningMean, resultRunningMean_ref);
-            pass = pass && ck::utils::check_err(resultRunningVariance, resultRunningVariance_ref);
+            pass = pass && ck::utils::check_err(resultRunningMean,
+                                                resultRunningMean_ref,
+                                                "Incorrect running mean values");
+            pass = pass && ck::utils::check_err(resultRunningVariance,
+                                                resultRunningVariance_ref,
+                                                "Incorrect running variance values");
         };
 
         if(saveMeanAndInvVariance)
@@ -438,8 +442,11 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
             resultSaveMean_dev.FromDevice(resultSaveMean.mData.data());
             resultSaveInvVariance_dev.FromDevice(resultSaveInvVariance.mData.data());
 
-            pass = pass && ck::utils::check_err(resultSaveMean, resultSaveMean_ref);
-            pass = pass && ck::utils::check_err(resultSaveInvVariance, resultSaveInvVariance_ref);
+            pass = pass && ck::utils::check_err(
+                               resultSaveMean, resultSaveMean_ref, "Incorrect saved mean values");
+            pass = pass && ck::utils::check_err(resultSaveInvVariance,
+                                                resultSaveInvVariance_ref,
+                                                "Incorrect saved invvariance values");
         };
     };
 
diff --git a/example/34_batchnorm/batchnorm_forward_training_nhwc_obsolete.cpp b/example/34_batchnorm/batchnorm_forward_training_nhwc_obsolete.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ffb9f4b58468699c1a34ff6aff101c2dd4adaf66
--- /dev/null
+++ b/example/34_batchnorm/batchnorm_forward_training_nhwc_obsolete.cpp
@@ -0,0 +1,598 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <limits>
+#include <iostream>
+#include <vector>
+#include <array>
+#include <algorithm>
+#include <getopt.h>
+
+#include "ck/ck.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+static struct option long_options[] = {{"inOutLengths", required_argument, nullptr, 'D'},
+                                       {"verify", required_argument, nullptr, 'v'},
+                                       {"help", no_argument, nullptr, '?'},
+                                       {nullptr, 0, nullptr, 0}};
+
+class BatchNormFwdArg
+{
+    private:
+    int option_index = 0;
+
+    public:
+    std::vector<size_t> inOutLengths;
+
+    bool do_verification = false;
+
+    bool updateMovingAverage;
+    bool saveMeanAndInvVariance;
+
+    int data_type               = 0;
+    int init_method             = 2;
+    bool time_kernel            = false;
+    bool use_multiblock_welford = false;
+
+    public:
+    void show_usage(const char* cmd)
+    {
+        std::cout << "Usage of " << cmd << std::endl;
+        std::cout << "--inOutLengths or -D, comma separated list of input tensor dimension "
+                     "lengths, must have 4 integers for nhwc"
+                  << std::endl;
+        std::cout << "--verify or -v, 1/0 to indicate whether to verify the batch-normalization "
+                     "result by "
+                     "comparing with the host-based batch-normalization"
+                  << std::endl;
+        std::cout << "Arg1: data type (0: fp16, 1: fp32, 3: int8, 5: bp16, 6: fp64)" << std::endl;
+        std::cout << "Arg2: 1/0 to indicate whether to update the moving average and variance "
+                     "(0=no, 1=yes)"
+                  << std::endl;
+        std::cout << "Arg3: 1/0 to indicate whether to save the calculated mean and invVariance "
+                     "(0=no, 1=yes)"
+                  << std::endl;
+        std::cout << "Arg4: init method used for bnScale and bnBias (0=no init, 1=single integer "
+                     "value, 2=scope integer "
+                     "value, 3=decimal value)"
+                  << std::endl;
+        std::cout << "Arg5: time kernel (0=no, 1=yes)" << std::endl;
+        std::cout << "Arg6: use multi-block welford (0=n0, 1=yes)" << std::endl;
+    };
+
+    int processArgs(int argc, char* argv[])
+    {
+        using ck::host_common::getTypeValuesFromString;
+
+        int ch;
+
+        while(1)
+        {
+            ch = getopt_long(argc, argv, "D:v:", long_options, &option_index);
+            if(ch == -1)
+                break;
+            switch(ch)
+            {
+            case 'D':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                inOutLengths = getTypeValuesFromString<size_t>(optarg);
+
+                if(inOutLengths.size() != 4)
+                    throw std::runtime_error(
+                        "NHWC tensor layout should have 4 length values specified!");
+                break;
+            case 'v':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                do_verification = static_cast<bool>(std::atoi(optarg));
+                break;
+            case '?':
+                if(std::string(long_options[option_index].name) == "help")
+                {
+                    show_usage(argv[0]);
+                    return (-1);
+                };
+                break;
+            default: show_usage(argv[0]); return (-1);
+            };
+        };
+
+        if(optind + 6 > argc)
+            throw std::runtime_error("Invalid cmd-line arguments, more argumetns are needed!");
+
+        data_type              = std::atoi(argv[optind++]);
+        updateMovingAverage    = std::atoi(argv[optind++]);
+        saveMeanAndInvVariance = std::atoi(argv[optind++]);
+        init_method            = std::atoi(argv[optind++]);
+        time_kernel            = static_cast<bool>(std::atoi(argv[optind++]));
+        use_multiblock_welford = static_cast<bool>(std::atoi(argv[optind]));
+
+        if(data_type != 0 && data_type != 1 && data_type != 3 && data_type != 5 && data_type != 6)
+            return (-1);
+
+        return (0);
+    };
+};
+
+using namespace ck;
+
+template <typename InOutDataType, typename AccDataType, bool UseMultiblockInK>
+bool bnorm_fwd_nhwc_test(bool do_verification,
+                         int init_method,
+                         bool time_kernel,
+                         const std::vector<size_t> inOutLengths,
+                         bool updateMovingAverage,
+                         bool saveMeanAndInvVariance,
+                         double averageFactor,
+                         double epsilon)
+{
+    // for NHWC BatchNorm calculation of mean and meansquare
+    constexpr int Rank         = 4;
+    constexpr int NumReduceDim = 3;
+
+    // when using lengths[] to create a tensor, lengths[0] is the length of highest dimension
+    // eg. N of NHWC, so lengths[3] is the dimension C length of NHWC
+    const std::vector<size_t> scaleBiasMeanVarLengths = {inOutLengths[3]};
+
+    // input data of the batchnorm forward algorithm
+    Tensor<InOutDataType> x(inOutLengths);
+    Tensor<AccDataType> bnScale(scaleBiasMeanVarLengths);
+    Tensor<AccDataType> bnBias(scaleBiasMeanVarLengths);
+
+    // output data of the batchnorm forward algorithm
+    Tensor<InOutDataType> y_ref(inOutLengths);
+    Tensor<InOutDataType> y(inOutLengths);
+
+    Tensor<AccDataType> resultSaveMean_ref(scaleBiasMeanVarLengths);
+    Tensor<AccDataType> resultSaveInvVariance_ref(scaleBiasMeanVarLengths);
+
+    Tensor<AccDataType> resultRunningMean_ref(scaleBiasMeanVarLengths);
+    Tensor<AccDataType> resultRunningVariance_ref(scaleBiasMeanVarLengths);
+
+    auto inOutStrides            = x.mDesc.GetStrides();
+    auto scaleBiasMeanVarStrides = bnScale.mDesc.GetStrides();
+
+    std::size_t num_thread = std::thread::hardware_concurrency();
+
+    if(updateMovingAverage)
+    {
+        if constexpr(std::is_same<InOutDataType, int8_t>::value)
+        {
+            x.GenerateTensorValue(GeneratorTensor_2<InOutDataType>{-5, 5}, num_thread);
+
+            const float x_mean       = 0.0f;
+            const float x_stddev     = 2.5f;
+            const float noise_stddev = 0.04f;
+
+            resultRunningMean_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_mean, noise_stddev}, num_thread);
+
+            resultRunningVariance_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_stddev * x_stddev, noise_stddev}, num_thread);
+        }
+        else
+        {
+            const float x_mean       = 0.0f;
+            const float x_stddev     = 1.0f;
+            const float noise_stddev = 0.04f;
+
+            // input data in normal distribution
+            x.GenerateTensorValue(GeneratorTensor_4<InOutDataType>{x_mean, x_stddev}, num_thread);
+
+            // initialize the runningMean to be values with tiny variation to the mean of the x
+            // values
+            resultRunningMean_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_mean, noise_stddev}, num_thread);
+
+            // initialize the runningVariance to be values with tiny variation to the variance of
+            // the x values
+            resultRunningVariance_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_stddev * x_stddev, noise_stddev}, num_thread);
+        };
+    }
+    else
+    {
+        if constexpr(std::is_same<InOutDataType, int8_t>::value)
+            x.GenerateTensorValue(GeneratorTensor_2<InOutDataType>{-5, 5}, num_thread);
+        else
+            x.GenerateTensorValue(GeneratorTensor_3<InOutDataType>{-5.0f, 5.0f}, num_thread);
+    };
+
+    if(do_verification)
+    {
+        switch(init_method)
+        {
+        case 0:
+            bnScale.GenerateTensorValue(GeneratorTensor_0<AccDataType>{}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_0<AccDataType>{}, num_thread);
+            break;
+        case 1:
+            bnScale.GenerateTensorValue(GeneratorTensor_1<AccDataType>{1}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_1<AccDataType>{0}, num_thread);
+            break;
+        case 2:
+            bnScale.GenerateTensorValue(GeneratorTensor_2<AccDataType>{-5, 5}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_2<AccDataType>{-5, 5}, num_thread);
+            break;
+        default:
+            bnScale.GenerateTensorValue(GeneratorTensor_3<AccDataType>{-5.0f, 5.0f}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_3<AccDataType>{-5.0f, 5.0f}, num_thread);
+        }
+    };
+
+    // these buffers are usually provided by the user application
+    DeviceMem x_dev(sizeof(InOutDataType) * x.mDesc.GetElementSpaceSize());
+    DeviceMem y_dev(sizeof(InOutDataType) * y.mDesc.GetElementSpaceSize());
+    DeviceMem bnScale_dev(sizeof(AccDataType) * bnScale.mDesc.GetElementSpaceSize());
+    DeviceMem bnBias_dev(sizeof(AccDataType) * bnBias.mDesc.GetElementSpaceSize());
+
+    // mean_dev or resultSaveMean_dev
+    DeviceMem resultSaveMean_dev(sizeof(AccDataType) *
+                                 resultSaveMean_ref.mDesc.GetElementSpaceSize());
+    // meansquare_dev or resultSaveInvVariance_dev
+    DeviceMem resultSaveInvVariance_dev(sizeof(AccDataType) *
+                                        resultSaveInvVariance_ref.mDesc.GetElementSpaceSize());
+    // resultRunningMean_dev
+    DeviceMem resultRunningMean_dev(sizeof(AccDataType) *
+                                    resultRunningMean_ref.mDesc.GetElementSpaceSize());
+    // resultRunningVariance_dev
+    DeviceMem resultRunningVariance_dev(sizeof(AccDataType) *
+                                        resultRunningVariance_ref.mDesc.GetElementSpaceSize());
+
+    x_dev.ToDevice(x.mData.data());
+    bnScale_dev.ToDevice(bnScale.mData.data());
+    bnBias_dev.ToDevice(bnBias.mData.data());
+
+    if(updateMovingAverage)
+    {
+        resultRunningMean_dev.ToDevice(resultRunningMean_ref.mData.data());
+        resultRunningVariance_dev.ToDevice(resultRunningVariance_ref.mData.data());
+    };
+
+    std::array<index_t, Rank> i_inOutLengths;
+    std::array<index_t, Rank> i_inOutStrides;
+    std::array<index_t, Rank - NumReduceDim> i_scaleBiasMeanVarLengths;
+    std::array<index_t, Rank - NumReduceDim> i_scaleBiasMeanVarStrides;
+
+    ck::ranges::copy(inOutLengths, i_inOutLengths.begin());
+    ck::ranges::copy(inOutStrides, i_inOutStrides.begin());
+    ck::ranges::copy(scaleBiasMeanVarLengths, i_scaleBiasMeanVarLengths.begin());
+    ck::ranges::copy(scaleBiasMeanVarStrides, i_scaleBiasMeanVarStrides.begin());
+
+    using PassThroughOp = ck::tensor_operation::element_wise::PassThrough;
+
+    using DeviceBatchNormFwdInstance =
+        ck::tensor_operation::device::DeviceBatchNormFwdImpl<InOutDataType,
+                                                             InOutDataType,
+                                                             AccDataType,
+                                                             AccDataType,   // ScaleDataType
+                                                             AccDataType,   // BiasDataType
+                                                             AccDataType,   // MeanVarDataType
+                                                             PassThroughOp, // YElementwiseOp
+                                                             Rank,
+                                                             NumReduceDim,
+                                                             UseMultiblockInK,
+                                                             256,
+                                                             16,
+                                                             16,
+                                                             1,
+                                                             2,
+                                                             0,
+                                                             1,
+                                                             1,
+                                                             1,
+                                                             1,
+                                                             1>;
+
+    auto batchnorm_fwd = DeviceBatchNormFwdInstance{};
+
+    auto argument_ptr = batchnorm_fwd.MakeArgumentPointer(
+        i_inOutLengths,
+        i_inOutStrides,
+        i_inOutStrides,
+        {0, 1, 2}, // indicates physical indices of reduce dimensions in lengths[] and strides[]
+        i_scaleBiasMeanVarLengths,
+        i_scaleBiasMeanVarStrides,
+        i_scaleBiasMeanVarStrides,
+        i_scaleBiasMeanVarStrides,
+        x_dev.GetDeviceBuffer(),
+        bnScale_dev.GetDeviceBuffer(),
+        bnBias_dev.GetDeviceBuffer(),
+        epsilon,
+        PassThroughOp{},
+        y_dev.GetDeviceBuffer(),
+        saveMeanAndInvVariance ? resultSaveMean_dev.GetDeviceBuffer() : nullptr,
+        saveMeanAndInvVariance ? resultSaveInvVariance_dev.GetDeviceBuffer() : nullptr,
+        averageFactor,
+        updateMovingAverage ? resultRunningMean_dev.GetDeviceBuffer() : nullptr,
+        updateMovingAverage ? resultRunningVariance_dev.GetDeviceBuffer() : nullptr);
+
+    if(!batchnorm_fwd.IsSupportedArgument(argument_ptr.get()))
+    {
+        std::cout << "The runtime parameters seems not supported by the BatchNorm device instance, "
+                     "exiting!"
+                  << std::endl;
+        return (false);
+    };
+
+    size_t workspace_sz = batchnorm_fwd.GetWorkSpaceSize(argument_ptr.get());
+
+    DeviceMem workspace_dev(workspace_sz);
+
+    batchnorm_fwd.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
+    auto invoker_ptr = batchnorm_fwd.MakeInvokerPointer();
+
+    if(time_kernel)
+    {
+        float avg_time   = 0.0f;
+        size_t num_bytes = 0;
+
+        size_t total_length = inOutLengths[0] * inOutLengths[1] * inOutLengths[2] * inOutLengths[3];
+        size_t invariant_length = inOutLengths[3];
+
+        avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        // inputing of x, scale, bias, outputing of y
+        num_bytes +=
+            total_length * sizeof(InOutDataType) * 2 + invariant_length * sizeof(AccDataType) * 2;
+
+        // outputing of mean, inv-variance
+        num_bytes += saveMeanAndInvVariance ? invariant_length * sizeof(AccDataType) * 2 : 0;
+
+        // updating of moving mean, variance
+        num_bytes += updateMovingAverage ? invariant_length * sizeof(AccDataType) * 4 : 0;
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        std::cout << "Perf: " << avg_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
+    }
+    else
+        (void)invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+
+        using ReferenceBatchNormFwdInstance =
+            ck::tensor_operation::host::ReferenceBatchNormFwd<InOutDataType,
+                                                              InOutDataType,
+                                                              AccDataType,
+                                                              AccDataType,
+                                                              AccDataType,
+                                                              AccDataType,
+                                                              PassThroughOp,
+                                                              Rank,
+                                                              NumReduceDim>;
+
+        auto batchNormFwd_ref = ReferenceBatchNormFwdInstance{};
+
+        auto argument_ptr_ref = batchNormFwd_ref.MakeArgumentPointer(
+            i_inOutLengths,
+            i_inOutStrides,
+            i_inOutStrides,
+            {0, 1, 2}, // indicates physical indices of reduce dimensions in lengths[] and strides[]
+            i_scaleBiasMeanVarLengths,
+            i_scaleBiasMeanVarStrides,
+            i_scaleBiasMeanVarStrides,
+            i_scaleBiasMeanVarStrides,
+            x.mData.data(),
+            bnScale.mData.data(),
+            bnBias.mData.data(),
+            epsilon,
+            PassThroughOp{},
+            y_ref.mData.data(),
+            saveMeanAndInvVariance ? resultSaveMean_ref.mData.data() : nullptr,
+            saveMeanAndInvVariance ? resultSaveInvVariance_ref.mData.data() : nullptr,
+            averageFactor,
+            updateMovingAverage ? resultRunningMean_ref.mData.data() : nullptr,
+            updateMovingAverage ? resultRunningVariance_ref.mData.data() : nullptr);
+
+        if(!batchNormFwd_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters seems not supported by the BatchNorm reference "
+                         "instance, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = batchNormFwd_ref.MakeInvokerPointer();
+
+        (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
+
+        y_dev.FromDevice(y.mData.data());
+        pass = pass && ck::utils::check_err(y, y_ref, "Incorrect normalized output values");
+
+        if(updateMovingAverage)
+        {
+            Tensor<AccDataType> resultRunningMean(scaleBiasMeanVarLengths);
+            Tensor<AccDataType> resultRunningVariance(scaleBiasMeanVarLengths);
+
+            resultRunningMean_dev.FromDevice(resultRunningMean.mData.data());
+            resultRunningVariance_dev.FromDevice(resultRunningVariance.mData.data());
+
+            pass = pass && ck::utils::check_err(resultRunningMean,
+                                                resultRunningMean_ref,
+                                                "Incorrect running mean values");
+            pass = pass && ck::utils::check_err(resultRunningVariance,
+                                                resultRunningVariance_ref,
+                                                "Incorrect running variance values");
+        };
+
+        if(saveMeanAndInvVariance)
+        {
+            using ck::host_common::dumpBufferToFile;
+
+            Tensor<AccDataType> resultSaveMean(scaleBiasMeanVarLengths);
+            Tensor<AccDataType> resultSaveInvVariance(scaleBiasMeanVarLengths);
+
+            resultSaveMean_dev.FromDevice(resultSaveMean.mData.data());
+            resultSaveInvVariance_dev.FromDevice(resultSaveInvVariance.mData.data());
+
+            pass = pass && ck::utils::check_err(
+                               resultSaveMean, resultSaveMean_ref, "Incorrect saved mean values");
+            pass = pass && ck::utils::check_err(resultSaveInvVariance,
+                                                resultSaveInvVariance_ref,
+                                                "Incorrect saved invvariance values");
+        };
+    };
+
+    return (pass);
+};
+
+const double epsilon              = std::numeric_limits<float>::epsilon();
+static const double averageFactor = 0.1;
+
+int main(int argc, char* argv[])
+{
+    bool pass = true;
+
+    if(argc > 1)
+    {
+        BatchNormFwdArg arg;
+
+        if(arg.processArgs(argc, argv) < 0)
+            return (-1);
+
+        if(arg.data_type == 0)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<ck::half_t, float, true>(arg.do_verification,
+                                                                    arg.init_method,
+                                                                    arg.time_kernel,
+                                                                    arg.inOutLengths,
+                                                                    arg.updateMovingAverage,
+                                                                    arg.saveMeanAndInvVariance,
+                                                                    averageFactor,
+                                                                    epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<ck::half_t, float, false>(arg.do_verification,
+                                                                     arg.init_method,
+                                                                     arg.time_kernel,
+                                                                     arg.inOutLengths,
+                                                                     arg.updateMovingAverage,
+                                                                     arg.saveMeanAndInvVariance,
+                                                                     averageFactor,
+                                                                     epsilon);
+        }
+        else if(arg.data_type == 1)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<float, float, true>(arg.do_verification,
+                                                               arg.init_method,
+                                                               arg.time_kernel,
+                                                               arg.inOutLengths,
+                                                               arg.updateMovingAverage,
+                                                               arg.saveMeanAndInvVariance,
+                                                               averageFactor,
+                                                               epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<float, float, false>(arg.do_verification,
+                                                                arg.init_method,
+                                                                arg.time_kernel,
+                                                                arg.inOutLengths,
+                                                                arg.updateMovingAverage,
+                                                                arg.saveMeanAndInvVariance,
+                                                                averageFactor,
+                                                                epsilon);
+        }
+        else if(arg.data_type == 3)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<int8_t, float, true>(arg.do_verification,
+                                                                arg.init_method,
+                                                                arg.time_kernel,
+                                                                arg.inOutLengths,
+                                                                arg.updateMovingAverage,
+                                                                arg.saveMeanAndInvVariance,
+                                                                averageFactor,
+                                                                epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<int8_t, float, false>(arg.do_verification,
+                                                                 arg.init_method,
+                                                                 arg.time_kernel,
+                                                                 arg.inOutLengths,
+                                                                 arg.updateMovingAverage,
+                                                                 arg.saveMeanAndInvVariance,
+                                                                 averageFactor,
+                                                                 epsilon);
+        }
+        else if(arg.data_type == 5)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<ck::bhalf_t, float, true>(arg.do_verification,
+                                                                     arg.init_method,
+                                                                     arg.time_kernel,
+                                                                     arg.inOutLengths,
+                                                                     arg.updateMovingAverage,
+                                                                     arg.saveMeanAndInvVariance,
+                                                                     averageFactor,
+                                                                     epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<ck::bhalf_t, float, false>(arg.do_verification,
+                                                                      arg.init_method,
+                                                                      arg.time_kernel,
+                                                                      arg.inOutLengths,
+                                                                      arg.updateMovingAverage,
+                                                                      arg.saveMeanAndInvVariance,
+                                                                      averageFactor,
+                                                                      epsilon);
+        }
+        else if(arg.data_type == 6)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<double, double, true>(arg.do_verification,
+                                                                 arg.init_method,
+                                                                 arg.time_kernel,
+                                                                 arg.inOutLengths,
+                                                                 arg.updateMovingAverage,
+                                                                 arg.saveMeanAndInvVariance,
+                                                                 averageFactor,
+                                                                 epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<double, double, false>(arg.do_verification,
+                                                                  arg.init_method,
+                                                                  arg.time_kernel,
+                                                                  arg.inOutLengths,
+                                                                  arg.updateMovingAverage,
+                                                                  arg.saveMeanAndInvVariance,
+                                                                  averageFactor,
+                                                                  epsilon);
+        }
+    }
+    else
+    {
+        pass = bnorm_fwd_nhwc_test<ck::half_t, float, true>(true,
+                                                            2,
+                                                            false, // don't time kernel
+                                                            {128, 16, 6, 512},
+                                                            true,
+                                                            true,
+                                                            averageFactor,
+                                                            epsilon);
+
+        pass = pass && bnorm_fwd_nhwc_test<ck::half_t, float, false>(true,
+                                                                     2,
+                                                                     false, // don't time kernel
+                                                                     {128, 16, 3, 1024},
+                                                                     true,
+                                                                     true,
+                                                                     averageFactor,
+                                                                     epsilon);
+    };
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/34_batchnorm/batchnorm_infer_impl.hpp b/example/34_batchnorm/batchnorm_infer_impl.hpp
index e457df81da2087ebf04b0e676964ec857b91ee77..d0b545b2a31d2fc095eece12ce273db823a5d2ac 100644
--- a/example/34_batchnorm/batchnorm_infer_impl.hpp
+++ b/example/34_batchnorm/batchnorm_infer_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -10,7 +10,7 @@
 #include "ck/utility/sequence.hpp"
 #include "ck/utility/tuple.hpp"
 #include "ck/utility/reduction_operator.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 
 #include "batchnorm_common.hpp"
 
@@ -46,7 +46,7 @@ int bnorm_infer(
     static_assert(NumBatchNormReduceDim < Rank,
                   "Invalid number of reduced dimensions for batchnorm!");
 
-    using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwise<
+    using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwiseImpl<
         ck::Tuple<XDataType, AccDataType, AccDataType, AccDataType, AccDataType>, // x, mean,
                                                                                   // variance,
                                                                                   // scale,
diff --git a/example/35_splitK_gemm/CMakeLists.txt b/example/35_splitK_gemm/CMakeLists.txt
index 794583954674f403a4902a6fc1cb244b803479d5..57ac33fc9a187d23a11560a8a60939ca8f818506 100644
--- a/example/35_splitK_gemm/CMakeLists.txt
+++ b/example/35_splitK_gemm/CMakeLists.txt
@@ -1,17 +1,23 @@
-add_custom_target(example_splitK_gemm_xdl)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(example_splitK_gemm_xdl)
+   add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
+   add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
+   add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
+   add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
 
-add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
-add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
-add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
-add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
-
-add_dependencies(example_splitK_gemm_xdl
+   add_dependencies(example_splitK_gemm_xdl
                  example_splitK_gemm_xdl_fp32
                  example_splitK_gemm_xdl_fp16
                  example_splitK_gemm_xdl_bfp16
                  example_splitK_gemm_xdl_int8)
 
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
-  add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
-endif()
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
+      add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
+   endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
index 7191ecf50ab5252b37855072b4e03f1194df52f6..1dc21a6c23ea69d60f0c51fd70b544899496afe8 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
index efdb315b4e5576aef8d992f6d57c231ac1f83a81..74fb16e15b019e145cdf13549254be1b09fbd8ca 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp
index bc2e3d1d52b668a2ea15c9f4ea1db44c4faeb19e..7506f694204b3c3aaa564704eda4f358baf17861 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp
index 4eb27824628d8f193ca6108a589794041fc2b2fb..7ebf9144082de53d64f3662e579dd6d246266f1d 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
index eefdbca6b1ae3e49cc221b3037e46888b4a9a9bd..0fc7a5cc29ccefdee6eef3f229715ee781e65223 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp b/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp
index f5eb4c3b6b0441451e9a6a667321b8d843dc36b8..d2337dcda5d30eb57e0e5f764f047cf21314a190 100644
--- a/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp
+++ b/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -9,7 +9,8 @@
 #include <ctime>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_sparse_embedding3_forward_layernorm.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
@@ -18,53 +19,26 @@
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp"
 
-// using EmbType       = float;
-// using IndexType     = int64_t;
-// using GammaDataType = float;
-// using BetaDataType  = float;
-// using AccDataType   = float;
-// using OutType       = float;
-
+// clang-format off
 using EmbType       = ck::half_t;
 using IndexType     = int64_t;
 using GammaDataType = ck::half_t;
 using BetaDataType  = ck::half_t;
 using AccDataType   = float;
 using OutType       = ck::half_t;
+using EmbElementwiseOperation = ck::tensor_operation::element_wise::AddAdd;
 
-// clang-format off
-//                                                                                                         BlockSize, DimClusterSize, RowClusterSize, DimPerBlock, RowPerBlock, DimThreadSize, RowVectorSize
-using DeviceInstance_fp32_e256   = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  256,   1, 1>;
-using DeviceInstance_fp32_e512   = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  512,   1, 1>;
-using DeviceInstance_fp32_e768   = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  768,   1, 1>;
-using DeviceInstance_fp32_e1024  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  1024,  1, 1>;
-using DeviceInstance_fp32_e1536  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  1536,  1, 1>;
-using DeviceInstance_fp32_e2048  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  2048,  1, 4>;
-using DeviceInstance_fp32_e4096  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  4096,  1, 4>;
-using DeviceInstance_fp32_e8192  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  8192,  1, 4>;
-using DeviceInstance_fp32_e16384 = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  16384, 1, 4>;
-
-using DeviceInstance_fp16_e256   = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  256,   1, 1>;
-using DeviceInstance_fp16_e512   = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  512,   1, 2>;
-using DeviceInstance_fp16_e768   = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  768,   1, 1>;
-using DeviceInstance_fp16_e1024  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  1024,  1, 2>;
-using DeviceInstance_fp16_e1536  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  1536,  1, 2>;
-using DeviceInstance_fp16_e2048  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  2048,  1, 2>;
-using DeviceInstance_fp16_e4096  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  4096,  1, 8>;
-using DeviceInstance_fp16_e8192  = ck::tensor_operation::device::DeviceSparseEmbedding3ForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, 256,  1,  256, 1,  8192,  1, 8>;
+using DeviceInstance_fp16_e256   = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  256,   1, 1, 3>;
+using DeviceInstance_fp16_e512   = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  512,   1, 2, 3>;
+using DeviceInstance_fp16_e768   = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  768,   1, 1, 3>;
+using DeviceInstance_fp16_e1024  = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  1024,  1, 2, 3>;
+using DeviceInstance_fp16_e1536  = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  1536,  1, 2, 3>;
+using DeviceInstance_fp16_e2048  = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  2048,  1, 2, 3>;
+using DeviceInstance_fp16_e4096  = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  4096,  1, 8, 3>;
+using DeviceInstance_fp16_e8192  = ck::tensor_operation::device::DeviceSparseEmbeddingsForwardLayernorm<EmbType, IndexType, GammaDataType, BetaDataType, AccDataType, OutType, EmbElementwiseOperation, 256,  1,  256, 1,  8192,  1, 8, 3>;
 
 template<typename emb_type, ck::index_t dim> struct emb_kernel{};
 
-template<> struct emb_kernel<float, 256>  { using kernel_type = DeviceInstance_fp32_e256; };
-template<> struct emb_kernel<float, 512>  { using kernel_type = DeviceInstance_fp32_e512; };
-template<> struct emb_kernel<float, 768>  { using kernel_type = DeviceInstance_fp32_e768; };
-template<> struct emb_kernel<float, 1024> { using kernel_type = DeviceInstance_fp32_e1024;};
-template<> struct emb_kernel<float, 1536> { using kernel_type = DeviceInstance_fp32_e1536;};
-template<> struct emb_kernel<float, 2048> { using kernel_type = DeviceInstance_fp32_e2048;};
-template<> struct emb_kernel<float, 4096> { using kernel_type = DeviceInstance_fp32_e4096;};
-template<> struct emb_kernel<float, 8192> { using kernel_type = DeviceInstance_fp32_e8192;};
-template<> struct emb_kernel<float, 16384>{ using kernel_type = DeviceInstance_fp32_e16384;};
-
 template<> struct emb_kernel<ck::half_t, 256>  { using kernel_type = DeviceInstance_fp16_e256; };
 template<> struct emb_kernel<ck::half_t, 512>  { using kernel_type = DeviceInstance_fp16_e512; };
 template<> struct emb_kernel<ck::half_t, 768>  { using kernel_type = DeviceInstance_fp16_e768; };
@@ -152,19 +126,20 @@ int main()
         beta_dev.ToDevice(beta.mData.data());
 
         auto device_instance = typename emb_kernel<EmbType, current_dim>::kernel_type{};
-        auto argument_ptr    = device_instance.MakeArgumentPointer(out_dev.GetDeviceBuffer(),
-                                                                emb_a_dev.GetDeviceBuffer(),
-                                                                emb_b_dev.GetDeviceBuffer(),
-                                                                emb_c_dev.GetDeviceBuffer(),
-                                                                index_a_dev.GetDeviceBuffer(),
-                                                                index_b_dev.GetDeviceBuffer(),
-                                                                index_c_dev.GetDeviceBuffer(),
-                                                                gamma_dev.GetDeviceBuffer(),
-                                                                beta_dev.GetDeviceBuffer(),
-                                                                num_rows,
-                                                                current_dim,
-                                                                index_length,
-                                                                epsilon);
+        auto argument_ptr    = device_instance.MakeArgumentPointer(
+            out_dev.GetDeviceBuffer(),
+            {ck::type_convert<EmbType*>(emb_a_dev.GetDeviceBuffer()),
+             ck::type_convert<EmbType*>(emb_b_dev.GetDeviceBuffer()),
+             ck::type_convert<EmbType*>(emb_c_dev.GetDeviceBuffer())},
+            {ck::type_convert<IndexType*>(index_a_dev.GetDeviceBuffer()),
+             ck::type_convert<IndexType*>(index_b_dev.GetDeviceBuffer()),
+             ck::type_convert<IndexType*>(index_c_dev.GetDeviceBuffer())},
+            gamma_dev.GetDeviceBuffer(),
+            beta_dev.GetDeviceBuffer(),
+            current_dim,
+            index_length,
+            epsilon,
+            EmbElementwiseOperation{});
         std::cout << "Dim:" << current_dim << ", kernel:" << device_instance.GetTypeString()
                   << std::endl
                   << std::flush;
diff --git a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
index 071e8a7431c057a32ea0061655d81a202774b923..b3d0ab6bf77aa85fa6b48fa503a3f3f0dde1bf12 100644
--- a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
+++ b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Computes C_m_o = Relu(A0[m, k] * B0[n, k] + D00[m, n] + D01[mn]) * B1[n, o] + D1[m, o]
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
index 9cf960c501ccb279febc8205505ca0edfdeefac4..de48093ac2b598e56afb31078bfecbceb10dc792 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+++ b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
@@ -1,7 +1,13 @@
-add_custom_target(example_grouped_conv_bwd_data)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(example_grouped_conv_bwd_data)
+   add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
+   add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
 
-add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
-add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
-
-add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
-add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
+   add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
+   add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/common.hpp b/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
index d07ee7bdc1c4b999f61a84727bb10fe7ea6f4c15..ca824b1075ced161760892fe6de53767b1079ccc 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
index 55ea8c3a3109469532d0e5c0566da3eb11e7353d..a3533bb4cc4b4874f0e7ae9f333347913d981ba0 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
index ddf82ec512c61c72a1a2e2323a6645f39f15ce6d..fb688b6f3f15383c2920a12e8902c2d4cf05a109 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc
index 0afd8bd70da849085db6e19583e933c3d1b91968..0f0b120cbcc578b2fa7f650c37c9a28148dd834a 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc
+++ b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 bool run_conv_bwd_data_bias_relu(const ExecutionConfig& config,
                                  const ck::utils::conv::ConvParam& conv_params,
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc
index e50c98bbe844f6ec438c781b7d5d8e254e57d98f..25678491ce90cdf1ab9455f14e091bd5fcd67431 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc
+++ b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 bool run_conv_bwd_data(const ExecutionConfig& config,
                        const ck::utils::conv::ConvParam& conv_params,
diff --git a/example/39_permute/common.hpp b/example/39_permute/common.hpp
index ab612cea1794c422c737a02e5bbe7a6728904c40..54f3a788097bebc6e293389cecd03d8ae68d8494 100644
--- a/example/39_permute/common.hpp
+++ b/example/39_permute/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/39_permute/permute_1xHxW_fp16.cpp b/example/39_permute/permute_1xHxW_fp16.cpp
index d7f9b80544a452ca7ea062d7d6135b92a2bf19ea..7336c3b631bcadd7d23a2d6b8e0b3d3a31787516 100644
--- a/example/39_permute/permute_1xHxW_fp16.cpp
+++ b/example/39_permute/permute_1xHxW_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/39_permute/permute_HxWx4_fp16.cpp b/example/39_permute/permute_HxWx4_fp16.cpp
index 342aa134ec5570c84447b137e5be165c9a5f697c..6c24919ded6a0c65556e305455861ca137203b3e 100644
--- a/example/39_permute/permute_HxWx4_fp16.cpp
+++ b/example/39_permute/permute_HxWx4_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/39_permute/permute_NxHxW_fp16.cpp b/example/39_permute/permute_NxHxW_fp16.cpp
index b53975eb2c8632583f61afde50a03eba32c17c9b..3551d2a7c8decf31a48ca628d99a53639f4fa1fe 100644
--- a/example/39_permute/permute_NxHxW_fp16.cpp
+++ b/example/39_permute/permute_NxHxW_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/39_permute/run_permute_bundle_example.inc b/example/39_permute/run_permute_bundle_example.inc
index 70406d63f91b95f4d1cce025051a74a0ee3d114e..2c198729226243fb1fb97db6beefa1fd1f62f625 100644
--- a/example/39_permute/run_permute_bundle_example.inc
+++ b/example/39_permute/run_permute_bundle_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/39_permute/run_permute_element_example.inc b/example/39_permute/run_permute_element_example.inc
index bc6235303039c7cb33a4057446137a9f18ddb186..35871344567ddf62a4fb3c2bd6a2e77cb94a1998 100644
--- a/example/39_permute/run_permute_element_example.inc
+++ b/example/39_permute/run_permute_element_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/40_conv2d_fwd_quantization/CMakeLists.txt b/example/40_conv2d_fwd_quantization/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b82a013b58f8e855615b494cf07bd1f03ce43556
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/CMakeLists.txt
@@ -0,0 +1,28 @@
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
+   add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
+   add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
+   add_example_executable(example_conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
+   set(target 1)
+ endif()
+endforeach()
+# Conv perlayer quantization
+add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
+
+# Conv perchannel quantization
+add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
+
+# Conv + bias + relu perlayer quantization
+add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
+
+# Conv + bias + relu perchannel quantization
+add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
+
+# Conv + bias + tanh perlayer quantization
+add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
+
+# Conv + bias + tanh perchannel quantization
+add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
\ No newline at end of file
diff --git a/example/40_conv2d_fwd_quantization/common.hpp b/example/40_conv2d_fwd_quantization/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..266b09145c84c456aedb2182554ced7815457410
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/common.hpp
@@ -0,0 +1,18 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..4573c68658bf44dc49ccade302feb1547746f0f9
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using BiasDataType         = int32_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename BiasLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType, RequantScaleDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout, RequantScaleLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_bias_perchannel_quantization_example(out_element_op);
+};
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..005f6263fd46e8b505fbc6859a5fa9ed334e7229
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
@@ -0,0 +1,84 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType   = int8_t;
+using WeiDataType  = int8_t;
+using BiasDataType = int32_t;
+using AccDataType  = int32_t;
+using OutDataType  = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename BiasLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_bias_perlayer_quantization_example(out_element_op);
+}
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..62e5e583de864524d2e937683a8c60cdcf9d57da
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
@@ -0,0 +1,87 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using BiasDataType         = int32_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::TanH;
+using OutElementOp =
+    ck::tensor_operation::element_wise::Add_Mul2_Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename BiasLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType, RequantScaleDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout, RequantScaleLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perchannel_quantization_example.inc"
+
+int main()
+{
+    float scale_z_inv         = 0.5f;
+    const auto out_element_op = OutElementOp{scale_z_inv, ActivationOp{}};
+    run_conv2d_fwd_bias_perchannel_quantization_example(out_element_op);
+};
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ef98fe7e4f945bab663b7e5ee895633186919df9
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType   = int8_t;
+using WeiDataType  = int8_t;
+using BiasDataType = int32_t;
+using AccDataType  = int32_t;
+using OutDataType  = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::TanH;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Mul_Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename BiasLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<BiasDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_bias_perlayer_quantization_example.inc"
+
+int main()
+{
+    float scale_acc           = 0.5f;
+    float scale_z_inv         = 0.5f;
+    const auto out_element_op = OutElementOp{scale_z_inv, scale_acc, ActivationOp{}};
+    run_conv2d_fwd_bias_perlayer_quantization_example(out_element_op);
+}
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e524ddb2b297ae218eaf197bea1f149e01a76efc
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<RequantScaleDataType>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<RequantScaleLayout>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_perchannel_quantization_example(out_element_op);
+}
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d29a3143c0ed4196a64ade05af0dc295a358c527
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
@@ -0,0 +1,79 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+using InDataType  = int8_t;
+using WeiDataType = int8_t;
+using AccDataType = int32_t;
+using OutDataType = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<
+        NDimSpatial,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<>,
+        OutDataType,
+        AccDataType,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<>,
+        OutLayout,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,            // ConvForwardSpecialization
+        GemmSpec,            // GemmSpecialization
+        256,                 // BlockSize
+        128,                 // MPerBlock
+        128,                 // NPerBlock
+        16,                  // K0PerBlock
+        4,                   // K1
+        4,                   // M1PerThread
+        4,                   // N1PerThread
+        1,                   // KPerThread
+        S<8, 2>,             // M1N1ThreadClusterM1Xs
+        S<8, 2>,             // M1N1ThreadClusterN1Xs
+        S<8, 1, 1, 4>,       // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+        S<2, 1, 128, 1>,     // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // ABlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+        S<1, 2, 0, 3>,       // ABlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+        S<8, 1, 1, 4>,       // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+        S<2, 1, 128, 1>,     // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferThreadClusterArrangeOrder
+        S<1, 2, 0, 3>,       // BBlockTransferSrcAccessOrder
+        S<4, 1, 1, 4>,       // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+        S<1, 2, 0, 3>,       // BBlockTransferSrcVectorTensorContiguousDimOrder
+        S<1, 1, 1, 4>,       // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+        S<0, 1, 2, 3, 4, 5>, // CThreadTransferSrcDstAccessOrder
+        5,                   // CThreadTransferSrcDstVectorDim
+        4>;                  // CThreadTransferDstScalarPerVector
+
+#include "run_conv2d_fwd_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_perlayer_quantization_example(out_element_op);
+}
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..06c839e4e226e082e9754f12ce39cececbd36269
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using BiasDataType         = int32_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using CShuffleDataType     = AccDataType;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename BiasLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout, RequantScaleLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<BiasDataType, RequantScaleDataType>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // BK1
+        32,          // MPerXdl
+        32,          // NPerXdl
+        2,           // MXdlPerWave
+        4,           // NXdlPerWave
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        8>;
+
+#include "run_conv2d_fwd_bias_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_bias_perchannel_quantization_example(out_element_op);
+};
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..7a9b42d39f971ab1c36476f4c0ead172cb6d2606
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
@@ -0,0 +1,88 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType       = int8_t;
+using WeiDataType      = int8_t;
+using BiasDataType     = int32_t;
+using AccDataType      = int32_t;
+using CShuffleDataType = AccDataType;
+using OutDataType      = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename BiasLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<BiasLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<BiasDataType>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // BK1
+        32,          // MPerXdl
+        32,          // NPerXdl
+        2,           // MXdlPerWave
+        4,           // NXdlPerWave
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        8>;
+
+#include "run_conv2d_fwd_bias_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_bias_perlayer_quantization_example(out_element_op);
+}
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..3495636297d5d83fb3ccc7d236e6135de9b715fd
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp
@@ -0,0 +1,87 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType           = int8_t;
+using WeiDataType          = int8_t;
+using RequantScaleDataType = float;
+using AccDataType          = int32_t;
+using CShuffleDataType     = AccDataType;
+using OutDataType          = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename RequantScaleLayout,
+          typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<RequantScaleLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<RequantScaleDataType>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // BK1
+        32,          // MPerXdl
+        32,          // NPerXdl
+        2,           // MXdlPerWave
+        4,           // NXdlPerWave
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        8>;
+
+#include "run_conv2d_fwd_perchannel_quantization_example.inc"
+
+int main()
+{
+    const auto out_element_op = OutElementOp{ActivationOp{}};
+    run_conv2d_fwd_perchannel_quantization_example(out_element_op);
+}
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..2611337254212b326f6eef540c6a4eca5cd19261
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+using InDataType       = int8_t;
+using WeiDataType      = int8_t;
+using AccDataType      = int32_t;
+using CShuffleDataType = AccDataType;
+using OutDataType      = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using ActivationOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        64,          // KPerBlock
+        16,          // AK1
+        16,          // BK1
+        32,          // MPerXdl
+        32,          // NPerXdl
+        2,           // MXdlPerWave
+        4,           // NXdlPerWave
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        16,          // ABlockTransferSrcScalarPerVector
+        16,          // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        16,          // BBlockTransferSrcScalarPerVector
+        16,          // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 64, 1, 4>,
+        16>;
+
+#include "run_conv2d_fwd_perlayer_quantization_example.inc"
+
+int main()
+{
+    float requant_scale       = 0.5f;
+    const auto out_element_op = OutElementOp{requant_scale, ActivationOp{}};
+    run_conv2d_fwd_perlayer_quantization_example(out_element_op);
+}
diff --git a/example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
similarity index 73%
rename from example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
rename to example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
index 832665edc055db434ce14044c15fbded8b51b847..e5b924ad5114a6e31a1ba2875118085329d5e8ce 100644
--- a/example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
@@ -1,97 +1,7 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/algorithm.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/utility/convolution_parameter.hpp"
-#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
-
-using InDataType           = int8_t;
-using WeiDataType          = int8_t;
-using BiasDataType         = int32_t;
-using RequantScaleDataType = float;
-using AccDataType          = int32_t;
-using CShuffleDataType     = int32_t;
-using OutDataType          = int8_t;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
-using InElementOp  = PassThrough;
-using WeiElementOp = PassThrough;
-using ActivationOp = ck::tensor_operation::element_wise::Relu;
-using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-template <ck::index_t NDimSpatial,
-          typename InLayout,
-          typename WeiLayout,
-          typename BiasLayout,
-          typename RequantScaleLayout,
-          typename OutLayout>
-using DeviceGroupedConvNDFwdInstance =
-    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
-        NDimSpatial,
-        InLayout,
-        WeiLayout,
-        ck::Tuple<BiasLayout, RequantScaleLayout>,
-        OutLayout,
-        InDataType,
-        WeiDataType,
-        AccDataType,
-        CShuffleDataType,
-        ck::Tuple<BiasDataType, RequantScaleDataType>,
-        OutDataType,
-        InElementOp,
-        WeiElementOp,
-        OutElementOp,
-        ConvSpec,    // ConvForwardSpecialization
-        GemmSpec,    // GemmSpecialization
-        1,           //
-        256,         // BlockSize
-        128,         // MPerBlock
-        256,         // NPerBlock
-        64,          // KPerBlock
-        16,          // AK1
-        16,          // BK1
-        32,          // MPerXdl
-        32,          // NPerXdl
-        2,           // MXdlPerWave
-        4,           // NXdlPerWave
-        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
-        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
-        2,           // ABlockTransferSrcVectorDim
-        16,          // ABlockTransferSrcScalarPerVector
-        16,          // ABlockTransferDstScalarPerVector_AK1
-        1,           // ABlockLdsExtraM
-        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
-        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
-        2,           // BBlockTransferSrcVectorDim
-        16,          // BBlockTransferSrcScalarPerVector
-        16,          // BBlockTransferDstScalarPerVector_BK1
-        1,           // BBlockLdsExtraN
-        1,
-        1,
-        S<1, 64, 1, 4>,
-        8>;
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#pragma once
 template <ck::index_t NDimSpatial,
           typename InDataType,
           typename WeiDataType,
@@ -219,12 +129,12 @@ bool run_grouped_conv_fwd(bool do_verification,
 
     if(do_verification)
     {
-        Tensor<CShuffleDataType> c_host(out_g_n_k_wos_desc);
+        Tensor<AccDataType> c_host(out_g_n_k_wos_desc);
 
         auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
                                                                      InDataType,
                                                                      WeiDataType,
-                                                                     CShuffleDataType,
+                                                                     AccDataType,
                                                                      InElementOp,
                                                                      WeiElementOp,
                                                                      PassThrough>();
@@ -257,7 +167,7 @@ bool run_grouped_conv_fwd(bool do_verification,
     return (pass ? 0 : 1);
 }
 
-int main()
+int run_conv2d_fwd_bias_perchannel_quantization_example(const OutElementOp& out_element_op)
 {
     bool do_verification           = true;
     bool time_kernel               = true;
@@ -268,7 +178,7 @@ int main()
         1,            // group
         4,            // batch
         64,           // output channels
-        32,           // input chanels
+        192,          // input chanels
         {3, 3},       // weight HW
         {71, 71},     // x HW
         {2, 2},       // strides
@@ -279,13 +189,12 @@ int main()
 
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{ActivationOp{}};
 
-    using InLayout           = ck::tensor_layout::convolution::GNHWC;
-    using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
+    using InLayout           = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout          = ck::tensor_layout::convolution::KYXGC;
     using BiasLayout         = ck::tensor_layout::convolution::G_K;
     using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-    using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+    using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
@@ -312,8 +221,6 @@ int main()
     const auto out_g_n_k_wos_desc =
         ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
 
-    std::cout << out_g_n_k_wos_desc << std::endl;
-
     using deviceOp = DeviceGroupedConvNDFwdInstance<ndim_spatial,
                                                     InLayout,
                                                     WeiLayout,
diff --git a/example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
similarity index 72%
rename from example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
rename to example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
index f5401350352ee0506f7fb99ceae1fcc52e22833a..9f3a769dcf0979d407169610f5cdbe15dbccb26e 100644
--- a/example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
@@ -1,94 +1,7 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/algorithm.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/utility/convolution_parameter.hpp"
-#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
-
-using InDataType       = int8_t;
-using WeiDataType      = int8_t;
-using BiasDataType     = int32_t;
-using AccDataType      = int32_t;
-using CShuffleDataType = int32_t;
-using OutDataType      = int8_t;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
-using InElementOp  = PassThrough;
-using WeiElementOp = PassThrough;
-using ActivationOp = ck::tensor_operation::element_wise::Relu;
-using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-template <ck::index_t NDimSpatial,
-          typename InLayout,
-          typename WeiLayout,
-          typename BiasLayout,
-          typename OutLayout>
-using DeviceGroupedConvNDFwdInstance =
-    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
-        NDimSpatial,
-        InLayout,
-        WeiLayout,
-        ck::Tuple<BiasLayout>,
-        OutLayout,
-        InDataType,
-        WeiDataType,
-        AccDataType,
-        CShuffleDataType,
-        ck::Tuple<BiasDataType>,
-        OutDataType,
-        InElementOp,
-        WeiElementOp,
-        OutElementOp,
-        ConvSpec,    // ConvForwardSpecialization
-        GemmSpec,    // GemmSpecialization
-        1,           //
-        256,         // BlockSize
-        128,         // MPerBlock
-        256,         // NPerBlock
-        64,          // KPerBlock
-        16,          // AK1
-        16,          // BK1
-        32,          // MPerXdl
-        32,          // NPerXdl
-        2,           // MXdlPerWave
-        4,           // NXdlPerWave
-        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
-        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
-        2,           // ABlockTransferSrcVectorDim
-        16,          // ABlockTransferSrcScalarPerVector
-        16,          // ABlockTransferDstScalarPerVector_AK1
-        1,           // ABlockLdsExtraM
-        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
-        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
-        2,           // BBlockTransferSrcVectorDim
-        16,          // BBlockTransferSrcScalarPerVector
-        16,          // BBlockTransferDstScalarPerVector_BK1
-        1,           // BBlockLdsExtraN
-        1,
-        1,
-        S<1, 64, 1, 4>,
-        8>;
+#pragma once
 
 template <ck::index_t NDimSpatial,
           typename InDataType,
@@ -205,12 +118,12 @@ bool run_grouped_conv_fwd(bool do_verification,
 
     if(do_verification)
     {
-        Tensor<CShuffleDataType> c_host(out_g_n_k_wos_desc);
+        Tensor<AccDataType> c_host(out_g_n_k_wos_desc);
 
         auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
                                                                      InDataType,
                                                                      WeiDataType,
-                                                                     CShuffleDataType,
+                                                                     AccDataType,
                                                                      InElementOp,
                                                                      WeiElementOp,
                                                                      PassThrough>();
@@ -242,7 +155,7 @@ bool run_grouped_conv_fwd(bool do_verification,
     return (pass ? 0 : 1);
 }
 
-int main()
+int run_conv2d_fwd_bias_perlayer_quantization_example(const OutElementOp& out_element_op)
 {
     bool do_verification           = true;
     bool time_kernel               = true;
@@ -253,7 +166,7 @@ int main()
         1,            // group
         4,            // batch
         64,           // output channels
-        32,           // input chanels
+        192,          // input chanels
         {3, 3},       // weight HW
         {71, 71},     // x HW
         {2, 2},       // strides
@@ -264,12 +177,11 @@ int main()
 
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{0.5f, ActivationOp{}};
 
-    using InLayout   = ck::tensor_layout::convolution::GNHWC;
-    using WeiLayout  = ck::tensor_layout::convolution::GKYXC;
+    using InLayout   = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout  = ck::tensor_layout::convolution::KYXGC;
     using BiasLayout = ck::tensor_layout::convolution::G_K;
-    using OutLayout  = ck::tensor_layout::convolution::GNHWK;
+    using OutLayout  = ck::tensor_layout::convolution::NHWGK;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
@@ -294,8 +206,6 @@ int main()
     const auto out_g_n_k_wos_desc =
         ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
 
-    std::cout << out_g_n_k_wos_desc << std::endl;
-
     return run_grouped_conv_fwd<
         ndim_spatial,
         InDataType,
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
new file mode 100644
index 0000000000000000000000000000000000000000..9b08fc690d8298b87ce3ebf89f462f3739734815
--- /dev/null
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
@@ -0,0 +1,234 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+template <ck::index_t NDimSpatial,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename InElementOp,
+          typename WeiElementOp,
+          typename OutElementOp,
+          typename DeviceConvNDFwdInstance>
+bool run_grouped_conv_fwd(bool do_verification,
+                          bool time_kernel,
+                          const ck::utils::conv::ConvParam& conv_param,
+                          const HostTensorDescriptor& in_g_n_c_wis_desc,
+                          const HostTensorDescriptor& wei_g_k_c_xs_desc,
+                          const HostTensorDescriptor& requant_scale_g_k_desc,
+                          const HostTensorDescriptor& out_g_n_k_wos_desc,
+                          const InElementOp& in_element_op,
+                          const WeiElementOp& wei_element_op,
+                          const OutElementOp& out_element_op)
+{
+    Tensor<InDataType> in(in_g_n_c_wis_desc);
+    Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<RequantScaleDataType> requant_scale(requant_scale_g_k_desc);
+    Tensor<OutDataType> out_host(out_g_n_k_wos_desc);
+    Tensor<OutDataType> out_device(out_g_n_k_wos_desc);
+
+    std::cout << "in: " << in.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "requant_scale: " << requant_scale.mDesc << std::endl;
+    std::cout << "out: " << out_host.mDesc << std::endl;
+
+    in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-128, 127});
+    wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-128, 127});
+    requant_scale.GenerateTensorValue(GeneratorTensor_2<RequantScaleDataType>{0, 1});
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem requant_scale_device_buf(sizeof(RequantScaleDataType) *
+                                       requant_scale.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+    requant_scale_device_buf.ToDevice(requant_scale.mData.data());
+
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> d0_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> d0_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), b_g_k_c_xs_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), b_g_k_c_xs_strides);
+    copy(requant_scale_g_k_desc.GetLengths(), d0_g_n_k_wos_lengths);
+    copy(requant_scale_g_k_desc.GetStrides(), d0_g_n_k_wos_strides);
+    copy(out_g_n_k_wos_desc.GetLengths(), e_g_n_k_wos_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), e_g_n_k_wos_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    // do Conv
+    auto conv     = DeviceConvNDFwdInstance{};
+    auto invoker  = conv.MakeInvoker();
+    auto argument = conv.MakeArgument(in_device_buf.GetDeviceBuffer(),
+                                      wei_device_buf.GetDeviceBuffer(),
+                                      {requant_scale_device_buf.GetDeviceBuffer()},
+                                      out_device_buf.GetDeviceBuffer(),
+                                      a_g_n_c_wis_lengths,
+                                      a_g_n_c_wis_strides,
+                                      b_g_k_c_xs_lengths,
+                                      b_g_k_c_xs_strides,
+                                      {d0_g_n_k_wos_lengths},
+                                      {d0_g_n_k_wos_strides},
+                                      e_g_n_k_wos_lengths,
+                                      e_g_n_k_wos_strides,
+                                      conv_filter_strides,
+                                      conv_filter_dilations,
+                                      input_left_pads,
+                                      input_right_pads,
+                                      in_element_op,
+                                      wei_element_op,
+                                      out_element_op);
+
+    if(!conv.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_conv with the specified compilation parameters does "
+            "not support this Conv problem");
+    }
+
+    float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop      = conv_param.GetFlops();
+    std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+    float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+    std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << conv.GetTypeString() << std::endl;
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+        Tensor<AccDataType> c_host(out_g_n_k_wos_desc);
+
+        auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
+                                                                     InDataType,
+                                                                     WeiDataType,
+                                                                     AccDataType,
+                                                                     InElementOp,
+                                                                     WeiElementOp,
+                                                                     PassThrough>();
+
+        auto ref_invoker  = ref_conv.MakeInvoker();
+        auto ref_argument = ref_conv.MakeArgument(in,
+                                                  wei,
+                                                  c_host,
+                                                  conv_param.conv_filter_strides_,
+                                                  conv_param.conv_filter_dilations_,
+                                                  conv_param.input_left_pads_,
+                                                  conv_param.input_right_pads_,
+                                                  in_element_op,
+                                                  wei_element_op,
+                                                  PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        // TODO: implement elementwise operation for host
+        out_host.ForEach([&](auto&, auto idx) {
+            out_element_op(out_host(idx), c_host(idx), requant_scale(idx));
+        });
+
+        out_device_buf.FromDevice(out_device.mData.data());
+
+        pass &=
+            ck::utils::check_err(out_device, out_host, "Error: incorrect results!", 1e-5f, 1e-4f);
+    }
+
+    return (pass ? 0 : 1);
+}
+
+int run_conv2d_fwd_perchannel_quantization_example(const OutElementOp& out_element_op)
+{
+    bool do_verification           = true;
+    bool time_kernel               = true;
+    const ck::index_t ndim_spatial = 2;
+
+    ck::utils::conv::ConvParam conv_param{
+        ndim_spatial, // n_dim
+        1,            // group
+        4,            // batch
+        64,           // output channels
+        192,          // input chanels
+        {3, 3},       // weight HW
+        {71, 71},     // x HW
+        {2, 2},       // strides
+        {1, 1},       // dilations
+        {1, 1},       // left_pads
+        {1, 1}        // right_pads
+    };
+
+    const auto in_element_op  = InElementOp{};
+    const auto wei_element_op = WeiElementOp{};
+
+    using InLayout           = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout          = ck::tensor_layout::convolution::KYXGC;
+    using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
+    using OutLayout          = ck::tensor_layout::convolution::NHWGK;
+
+    const auto in_g_n_c_wis_desc =
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
+
+    const auto wei_g_k_c_xs_desc =
+        ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(conv_param);
+
+    const auto requant_scale_g_k_desc =
+        HostTensorDescriptor({conv_param.G_,
+                              conv_param.N_,
+                              conv_param.K_,
+                              conv_param.output_spatial_lengths_[0],
+                              conv_param.output_spatial_lengths_[1]},
+                             {
+                                 conv_param.K_, // g
+                                 0,             // n
+                                 1,             // k
+                                 0,             // ho
+                                 0              // wo
+                             });
+
+    const auto out_g_n_k_wos_desc =
+        ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
+
+    using deviceOp = DeviceGroupedConvNDFwdInstance<ndim_spatial,
+                                                    InLayout,
+                                                    WeiLayout,
+                                                    RequantScaleLayout,
+                                                    OutLayout>;
+
+    return run_grouped_conv_fwd<ndim_spatial,
+                                InDataType,
+                                WeiDataType,
+                                OutDataType,
+                                InElementOp,
+                                WeiElementOp,
+                                OutElementOp,
+                                deviceOp>(do_verification,
+                                          time_kernel,
+                                          conv_param,
+                                          in_g_n_c_wis_desc,
+                                          wei_g_k_c_xs_desc,
+                                          requant_scale_g_k_desc,
+                                          out_g_n_k_wos_desc,
+                                          in_element_op,
+                                          wei_element_op,
+                                          out_element_op);
+}
diff --git a/example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
similarity index 68%
rename from example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
rename to example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
index 2d46d86655cdbaa6afc07a10dc50a5ce2a809bfe..267c737e00fc5046e06551ca3e366d01dc78d230 100644
--- a/example/44_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
@@ -1,89 +1,7 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/algorithm.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/utility/convolution_parameter.hpp"
-#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
-
-using InDataType       = int8_t;
-using WeiDataType      = int8_t;
-using AccDataType      = int32_t;
-using CShuffleDataType = int32_t;
-using OutDataType      = int8_t;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
-using InElementOp  = PassThrough;
-using WeiElementOp = PassThrough;
-using ActivationOp = PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
-using DeviceGroupedConvNDFwdInstance =
-    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
-        NDimSpatial,
-        InLayout,
-        WeiLayout,
-        ck::Tuple<>,
-        OutLayout,
-        InDataType,
-        WeiDataType,
-        AccDataType,
-        CShuffleDataType,
-        ck::Tuple<>,
-        OutDataType,
-        InElementOp,
-        WeiElementOp,
-        OutElementOp,
-        ConvSpec,    // ConvForwardSpecialization
-        GemmSpec,    // GemmSpecialization
-        1,           //
-        256,         // BlockSize
-        128,         // MPerBlock
-        256,         // NPerBlock
-        64,          // KPerBlock
-        16,          // AK1
-        16,          // BK1
-        32,          // MPerXdl
-        32,          // NPerXdl
-        2,           // MXdlPerWave
-        4,           // NXdlPerWave
-        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
-        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
-        2,           // ABlockTransferSrcVectorDim
-        16,          // ABlockTransferSrcScalarPerVector
-        16,          // ABlockTransferDstScalarPerVector_AK1
-        1,           // ABlockLdsExtraM
-        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
-        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
-        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
-        2,           // BBlockTransferSrcVectorDim
-        16,          // BBlockTransferSrcScalarPerVector
-        16,          // BBlockTransferDstScalarPerVector_BK1
-        1,           // BBlockLdsExtraN
-        1,
-        1,
-        S<1, 64, 1, 4>,
-        16>;
+#pragma once
 
 template <ck::index_t NDimSpatial,
           typename InDataType,
@@ -221,10 +139,10 @@ bool run_grouped_conv_fwd(bool do_verification,
     return (pass ? 0 : 1);
 }
 
-int main()
+int run_conv2d_fwd_perlayer_quantization_example(const OutElementOp& out_element_op)
 {
     bool do_verification           = true;
-    bool time_kernel               = true;
+    bool time_kernel               = false;
     const ck::index_t ndim_spatial = 2;
 
     ck::utils::conv::ConvParam conv_param{
@@ -232,7 +150,7 @@ int main()
         1,            // group
         4,            // batch
         64,           // output channels
-        32,           // input chanels
+        192,          // input chanels
         {3, 3},       // weight HW
         {71, 71},     // x HW
         {2, 2},       // strides
@@ -243,11 +161,10 @@ int main()
 
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{0.5f, ActivationOp{}};
 
-    using InLayout  = ck::tensor_layout::convolution::GNHWC;
-    using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-    using OutLayout = ck::tensor_layout::convolution::GNHWK;
+    using InLayout  = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout = ck::tensor_layout::convolution::KYXGC;
+    using OutLayout = ck::tensor_layout::convolution::NHWGK;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
diff --git a/example/41_grouped_conv_conv_fwd/CMakeLists.txt b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
index 9cb30f61760a47a2f72c98e31de2b522b97543ca..ae251e88d205a04968eab430b52c004a1136d853 100644
--- a/example/41_grouped_conv_conv_fwd/CMakeLists.txt
+++ b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
@@ -1,8 +1,18 @@
-add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list2 gfx908 gfx90a)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
+   endif(USE_BITINT_EXTENSION_INT4)
+   set(target 1)
+ endif()
+endforeach()
 
-if(USE_BITINT_EXTENSION_INT4)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
-endif(USE_BITINT_EXTENSION_INT4)
+if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+endif()
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp
index 2aea08c40004fac22ea0e45521ad09fd806751d8..e37d41369599819bdf1b9836c1a0587be2e31a0f 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp
index b7f80e76d6ce5a690beef4f48361e0a65402cb47..496e676a402f4d12a325c2afef56ec422cb4d721 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp
index 15e460948ef528b697a4e3bad7d40ca332f85c15..35d50721dcea27b2bfb42eea6e39da08709dfc7d 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp
index 2cc4c07c0d89fc83b233b50967cd595dc0e53ac7..80f6e9ae05712b1df4b05b1bb75fbff62d9ca70a 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp
index 40ff0f69cc14cd644361f3391eea8ed52380adc0..3ade6c811ae5583bbb8a2a3b277c6f4e7f265a87 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc b/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc
index a2c97f4d421f9382aa279b257a49299c4016e2c3..0722d497d8df0ebf0acf0cf798215bebcdee6e3a 100644
--- a/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc
+++ b/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/42_groupnorm/CMakeLists.txt b/example/42_groupnorm/CMakeLists.txt
index c3b7b825920314aba3fa15e401da26860c23a3c3..e8c306ac582f56607c8bfa309c4a3f16cd64be50 100644
--- a/example/42_groupnorm/CMakeLists.txt
+++ b/example/42_groupnorm/CMakeLists.txt
@@ -1 +1,3 @@
-add_example_executable(example_groupnorm_sigmoid_fp16 groupnorm_sigmoid_fp16.cpp)
+add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
+add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
+add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
diff --git a/example/42_groupnorm/common.hpp b/example/42_groupnorm/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..c8f91eb53b5b943bfbdd264b842cf7762a176b82
--- /dev/null
+++ b/example/42_groupnorm/common.hpp
@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <getopt.h>
+
+#include "ck/ck.hpp"
+#include "ck/utility/reduction_enums.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+
+#include "ck/library/utility/fill.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp"
diff --git a/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..cc107b63dcd87735ba40ebbe7633c61993406166
--- /dev/null
+++ b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
@@ -0,0 +1,56 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+constexpr int Rank         = 5;
+constexpr int NumReduceDim = 3;
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+
+struct YElementOp
+{
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(ck::is_same<T, float>::value || ck::is_same<T, double>::value ||
+                          ck::is_same<T, ck::half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        T a;
+
+        ck::tensor_operation::element_wise::Sigmoid{}(a, x);
+
+        y = x * a;
+    };
+};
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
+                                                          GammaDataType,
+                                                          BetaDataType,
+                                                          ComputeDataType,
+                                                          YDataType,
+                                                          YElementOp,
+                                                          Rank,
+                                                          NumReduceDim,
+                                                          1024, // BlockSize
+                                                          1,    // ClusterM
+                                                          1024, // ClusterK
+                                                          1,    // SliceM
+                                                          32,   // SliceK
+                                                          1,    // SrcVecDim (0=M, 1=K)
+                                                          2,    // SrcScalarPerVector
+                                                          1,    // GammaVecDim (0=M, 1=K)
+                                                          2,    // GammaScalarPerVector
+                                                          1,    // BetaVecDim (0=M, 1=K)
+                                                          2,    // BetaScalarPerVector
+                                                          2>;   // OutScalarPerVector
+
+#include "run_groupnorm_example.inc"
+
+int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
diff --git a/example/42_groupnorm/groupnorm_splitk_fp16.cpp b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..057b240a63fc38217c095b6f3a058e1d4db6ea27
--- /dev/null
+++ b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+constexpr int Rank         = 5;
+constexpr int NumReduceDim = 3;
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using YElementOp      = ck::tensor_operation::element_wise::Swish;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
+                                                                GammaDataType,
+                                                                BetaDataType,
+                                                                ComputeDataType,
+                                                                YDataType,
+                                                                YElementOp,
+                                                                Rank,
+                                                                NumReduceDim,
+                                                                256, // BlockSize
+                                                                1,   // ClusterM
+                                                                256, // ClusterK
+                                                                1,   // SliceM
+                                                                16,  // SliceK
+                                                                1,   // SrcVecDim (0=M, 1=K)
+                                                                2,   // SrcScalarPerVector
+                                                                1,   // GammaVecDim (0=M, 1=K)
+                                                                2,   // GammaScalarPerVector
+                                                                1,   // BetaVecDim (0=M, 1=K)
+                                                                2,   // BetaScalarPerVector
+                                                                2>;  // OutScalarPerVector
+
+#include "run_groupnorm_example.inc"
+
+int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
diff --git a/example/42_groupnorm/groupnorm_swish_fp16.cpp b/example/42_groupnorm/groupnorm_swish_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..363f22ed4c0015c9a96aa98a8b6b5302978e3d25
--- /dev/null
+++ b/example/42_groupnorm/groupnorm_swish_fp16.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+constexpr int Rank         = 5;
+constexpr int NumReduceDim = 3;
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using YElementOp      = ck::tensor_operation::element_wise::Swish;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
+                                                          GammaDataType,
+                                                          BetaDataType,
+                                                          ComputeDataType,
+                                                          YDataType,
+                                                          YElementOp,
+                                                          Rank,
+                                                          NumReduceDim,
+                                                          1024, // BlockSize
+                                                          1,    // ClusterM
+                                                          1024, // ClusterK
+                                                          1,    // SliceM
+                                                          32,   // SliceK
+                                                          1,    // SrcVecDim (0=M, 1=K)
+                                                          2,    // SrcScalarPerVector
+                                                          1,    // GammaVecDim (0=M, 1=K)
+                                                          2,    // GammaScalarPerVector
+                                                          1,    // BetaVecDim (0=M, 1=K)
+                                                          2,    // BetaScalarPerVector
+                                                          2>;   // OutScalarPerVector
+
+#include "run_groupnorm_example.inc"
+
+int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
diff --git a/example/42_groupnorm/groupnorm_sigmoid_fp16.cpp b/example/42_groupnorm/run_groupnorm_example.inc
similarity index 53%
rename from example/42_groupnorm/groupnorm_sigmoid_fp16.cpp
rename to example/42_groupnorm/run_groupnorm_example.inc
index e62001d6692d6dd504078cd12639167641f3cc24..16065c8d46a42433416e21e9607dc3fcb708d817 100644
--- a/example/42_groupnorm/groupnorm_sigmoid_fp16.cpp
+++ b/example/42_groupnorm/run_groupnorm_example.inc
@@ -1,80 +1,15 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-#include <getopt.h>
-
-#include "ck/ck.hpp"
-#include "ck/utility/reduction_enums.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
-#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
-
-#include "ck/library/utility/fill.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp"
-
-constexpr int Rank         = 5;
-constexpr int NumReduceDim = 3;
-
-using XDataType     = ck::half_t;
-using GammaDataType = ck::half_t;
-using BetaDataType  = ck::half_t;
-using YDataType     = ck::half_t;
-using AccDataType   = float;
-
-struct YElementOp
-{
-    template <typename T>
-    __host__ __device__ void operator()(T& y, const T& x) const
-    {
-        static_assert(ck::is_same<T, float>::value || ck::is_same<T, double>::value ||
-                          ck::is_same<T, ck::half_t>::value,
-                      "Data type is not supported by this operation!");
-
-        T a;
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-        ck::tensor_operation::element_wise::Sigmoid{}(a, x);
+#pragma once
 
-        y = x * a;
-    };
-};
-
-using DeviceInstance =
-    ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
-                                                          GammaDataType,
-                                                          BetaDataType,
-                                                          AccDataType,
-                                                          YDataType,
-                                                          YElementOp,
-                                                          Rank,
-                                                          NumReduceDim,
-                                                          1024, // BlockSize
-                                                          1,    // ClusterM
-                                                          1024, // ClusterK
-                                                          1,    // SliceM
-                                                          32,   // SliceK
-                                                          1,    // SrcVecDim (0=M, 1=K)
-                                                          2,    // SrcScalarPerVector
-                                                          1,    // GammaVecDim (0=M, 1=K)
-                                                          2,    // GammaScalarPerVector
-                                                          1,    // BetaVecDim (0=M, 1=K)
-                                                          2,    // BetaScalarPerVector
-                                                          2>;   // OutScalarPerVector
-
-int main(int argc, char* argv[])
+int run_groupnorm_example(int argc, char* argv[])
 {
-    ck::index_t N = 2;
-    ck::index_t H = 32;
-    ck::index_t W = 32;
-    ck::index_t G = 32;
-    ck::index_t C = 30;
+    ck::index_t N = 32;
+    ck::index_t H = 16;
+    ck::index_t W = 16;
+    ck::index_t G = 64;
+    ck::index_t C = 128;
 
     if(argc == 1)
     {
@@ -138,6 +73,10 @@ int main(int argc, char* argv[])
         return 1;
     };
 
+    size_t workspace_sz = device_instance.GetWorkSpaceSize(argument_ptr.get());
+    DeviceMem workspace_dev(workspace_sz);
+    device_instance.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
     auto invoker_ptr = device_instance.MakeInvokerPointer();
     float ave_time   = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true, true});
 
@@ -157,7 +96,7 @@ int main(int argc, char* argv[])
                                                                                  GammaDataType,
                                                                                  BetaDataType,
                                                                                  YDataType,
-                                                                                 AccDataType,
+                                                                                 ComputeDataType,
                                                                                  YElementOp>;
 
         ReferenceInstance ref;
diff --git a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
index 7ac4b68272e2b0a1a5cbc09fa3734e190d6897da..ebba88cf41fc493103dc78b1252e6d3ad5e19739 100644
--- a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
+++ b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -8,7 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
diff --git a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
index 764e55ef558bd977fca8f25f16b2bba4c111114b..4ab26293ccfddeb5117d409d0af9c00a303ce984 100644
--- a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
+++ b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -8,7 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
diff --git a/example/44_conv2d_fwd_quantization/CMakeLists.txt b/example/44_conv2d_fwd_quantization/CMakeLists.txt
deleted file mode 100644
index f02e5110d071e3a43b057c5a7c4b927373c1646b..0000000000000000000000000000000000000000
--- a/example/44_conv2d_fwd_quantization/CMakeLists.txt
+++ /dev/null
@@ -1,3 +0,0 @@
-add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
-add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
-add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
diff --git a/example/44_elementwise_permute/elementwise_permute_4D_fp16.cpp b/example/44_elementwise_permute/elementwise_permute_4D_fp16.cpp
index 0bbdbe52b9a7d813ec96cd18301e785a4d4e075a..2ceda86839abc37bd71fc1179d12be51d42e9a20 100644
--- a/example/44_elementwise_permute/elementwise_permute_4D_fp16.cpp
+++ b/example/44_elementwise_permute/elementwise_permute_4D_fp16.cpp
@@ -3,7 +3,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
@@ -19,13 +19,13 @@ using BDataType = F16;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 using DeviceElementwisePermuteInstance =
-    ck::tensor_operation::device::DeviceElementwise<ck::Tuple<ADataType>,
-                                                    ck::Tuple<BDataType>,
-                                                    PassThrough,
-                                                    4,
-                                                    8,
-                                                    ck::Sequence<8>,
-                                                    ck::Sequence<1>>;
+    ck::tensor_operation::device::DeviceElementwiseImpl<ck::Tuple<ADataType>,
+                                                        ck::Tuple<BDataType>,
+                                                        PassThrough,
+                                                        4,
+                                                        8,
+                                                        ck::Sequence<8>,
+                                                        ck::Sequence<1>>;
 
 template <typename HostTensorA, typename HostTensorB, typename Functor>
 void host_elementwise4D(HostTensorB& B_nhwc, const HostTensorA& A_nchw, Functor functor)
diff --git a/example/44_elementwise_permute/elementwise_permute_4D_fp16_2d.cpp b/example/44_elementwise_permute/elementwise_permute_4D_fp16_2d.cpp
index f16ad3b3c5c4c5b3aa7f53bb63c7a0149dd7584c..6b94a5d46f5ad20474900365ab6397ef48d87769 100644
--- a/example/44_elementwise_permute/elementwise_permute_4D_fp16_2d.cpp
+++ b/example/44_elementwise_permute/elementwise_permute_4D_fp16_2d.cpp
@@ -3,7 +3,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
-#include "ck/tensor_operation/gpu/device/device_elementwise_2d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
@@ -17,15 +17,15 @@ using BDataType = F16;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 using DeviceElementwisePermuteInstance =
-    ck::tensor_operation::device::DeviceElementwise<ck::Tuple<ADataType>,
-                                                    ck::Tuple<BDataType>,
-                                                    PassThrough,
-                                                    3, // NumDim_M
-                                                    1, // NumDim_N
-                                                    8,
-                                                    8,
-                                                    ck::Sequence<8>,
-                                                    ck::Sequence<8>>;
+    ck::tensor_operation::device::DeviceElementwise2dImpl<ck::Tuple<ADataType>,
+                                                          ck::Tuple<BDataType>,
+                                                          PassThrough,
+                                                          3, // NumDim_M
+                                                          1, // NumDim_N
+                                                          8,
+                                                          8,
+                                                          ck::Sequence<8>,
+                                                          ck::Sequence<8>>;
 
 template <typename HostTensorA, typename HostTensorB, typename Functor>
 void host_elementwise4D(HostTensorB& B_nhwc,
diff --git a/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp b/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
index 7d6ff12eeafe8de6f45e93567959d86d748c783c..76361f87a5b58071b149cab505f7c0485abfadab 100644
--- a/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
+++ b/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/46_gemm_add_multiply/CMakeLists.txt b/example/46_gemm_add_multiply/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..bfe057e8da677f30844a742b032522a2a61a947c
--- /dev/null
+++ b/example/46_gemm_add_multiply/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
+add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
diff --git a/example/46_gemm_add_multiply/README.md b/example/46_gemm_add_multiply/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..ee5cdee3659e7b6c3ba9bc083c48f91fcd66e581
--- /dev/null
+++ b/example/46_gemm_add_multiply/README.md
@@ -0,0 +1,26 @@
+# Instructions for ```example_gemm_add_multiply_dl_fp16```
+
+## Run ```example_gemm_add_multiply_dl_fp16```
+```bash
+#arg1: verification (0=no, 1=yes)
+#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
+#arg3: time kernel (0=no, 1=yes)
+#arg4 to 11: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD0, StrideD1, StrideE"
+./bin/example_gemm_add_multiply_dl_fp16 1 1 1
+```
+
+Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
+```
+a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
+b_k_n: dim 2, lengths {4096, 4096}, strides {4096, 1}
+d0_m_n: dim 2, lengths {3840, 4096}, strides {0, 1}
+d1_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
+e_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
+arg.a_grid_desc_k0_m0_m1_k1_{2048, 3840, 2}
+arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
+arg.e_grid_desc_m_n_{ 3840, 4096}
+launch_and_time_kernel: grid_dim {960, 1, 1}, block_dim {256, 1, 1}
+Warm up 1 time
+Start running 10 times...
+Perf: 3.99904 ms, 32.22 TFlops, 31.9913 GB/s, DeviceGemmMultipleD_Dl<256, 128, 128, 16, 2, 4, 4, 1>
+```
diff --git a/example/46_gemm_add_multiply/common.hpp b/example/46_gemm_add_multiply/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..2c656cf44159a594f0b8ba3b40b7d8d936d57cb2
--- /dev/null
+++ b/example/46_gemm_add_multiply/common.hpp
@@ -0,0 +1,102 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <algorithm>
+#include <cstddef>
+#include <iostream>
+#include <stdexcept>
+#include <string>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddMultiply = ck::tensor_operation::element_wise::AddMultiply;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+using I8   = int8_t;
+using I32  = int32_t;
+
+struct ProblemSize final
+{
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA  = 4096;
+    ck::index_t StrideB  = 4096;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideD1 = 4096;
+    ck::index_t StrideE  = 4096;
+};
+
+struct ExecutionConfig final
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+};
+
+inline bool
+parse_cmd_args(int argc, char* argv[], ProblemSize& problem_size, ExecutionConfig& config)
+{
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 12)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+
+        problem_size.M = std::stoi(argv[4]);
+        problem_size.N = std::stoi(argv[5]);
+        problem_size.K = std::stoi(argv[6]);
+
+        problem_size.StrideA  = std::stoi(argv[7]);
+        problem_size.StrideB  = std::stoi(argv[8]);
+        problem_size.StrideD0 = std::stoi(argv[9]);
+        problem_size.StrideD1 = std::stoi(argv[10]);
+        problem_size.StrideE  = std::stoi(argv[11]);
+    }
+    else
+    {
+        std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl
+                  << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
+                  << std::endl
+                  << "arg3: time kernel (0=no, 1=yes)" << std::endl
+                  << "arg4 to 10: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD0, StrideD1, "
+                     "StrideE"
+                  << std::endl;
+        return false;
+    }
+
+    return true;
+}
diff --git a/example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp b/example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..58a399f226c1b59073dd000e382e673badb5e1fb
--- /dev/null
+++ b/example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp"
+
+using ADataType   = F16;
+using BDataType   = F16;
+using AccDataType = F32;
+using D0DataType  = F16;
+using D1DataType  = F16;
+using DsDataType  = ck::Tuple<D0DataType, D1DataType>;
+using EDataType   = F16;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using D0Layout = Row;
+using D1Layout = Row;
+using DsLayout = ck::Tuple<D0Layout, D1Layout>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddMultiply;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// clang-format off
+using DeviceOpInstance = ck::tensor_operation::device::
+        //  ##################| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|     DsData|     EData|           A|           B|          CDE|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        //  ##################|        |        |         |        |      Type|      Type|        Type|       Type|      Type| Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        //  ##################|        |        |         |        |          |          |            |           |          |   Operation|   Operation|    Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        //  ##################|        |        |         |        |          |          |            |           |          |            |            |             |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        DeviceGemmMultipleD_Dl< ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                        BDataType,
+                                                                        AccDataType,
+                                                                        AccDataType,
+                                                                        AElementOp,
+                                                                        BElementOp,
+                                                                        PassThrough>;
+
+#include "run_gemm_add_multiply_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_add_multiply_example(argc, argv); }
diff --git a/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp b/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..56417b101d9528674535ab93f6fdc7ab36d47c3d
--- /dev/null
+++ b/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+
+using ADataType   = F16;
+using BDataType   = F16;
+using AccDataType = F32;
+using D0DataType  = F16;
+using D1DataType  = F16;
+using DsDataType  = ck::Tuple<D0DataType, D1DataType>;
+using EDataType   = F16;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using D0Layout = Row;
+using D1Layout = Row;
+using DsLayout = ck::Tuple<D0Layout, D1Layout>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddMultiply;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// clang-format off
+using DeviceOpInstance = ck::tensor_operation::device::
+        //##############################|      A|      B|       Ds|      E| AData| BData| AccData| CShuffle|     DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|   Layout| Layout|  Type|  Type|    Type| DataType|       Type|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |         |       |      |      |        |         |           |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |         |       |      |      |        |         |           |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, DsLayout,    Row,   F16,   F16,     F32,      F16, DsDataType,   F16, PassThrough, PassThrough, CDEElementOp,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                        BDataType,
+                                                                        AccDataType,
+                                                                        AccDataType,
+                                                                        AElementOp,
+                                                                        BElementOp,
+                                                                        PassThrough>;
+
+#include "run_gemm_add_multiply_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_add_multiply_example(argc, argv); }
diff --git a/example/46_gemm_add_multiply/run_gemm_add_multiply_example.inc b/example/46_gemm_add_multiply/run_gemm_add_multiply_example.inc
new file mode 100644
index 0000000000000000000000000000000000000000..e1b2bccfe110407b917ce29e12d3e9ea45301a5c
--- /dev/null
+++ b/example/46_gemm_add_multiply/run_gemm_add_multiply_example.inc
@@ -0,0 +1,139 @@
+#pragma once
+
+bool run_gemm_add_multiply(const ProblemSize& problem_size, const ExecutionConfig& config)
+{
+    using namespace ck::literals;
+
+    auto& [M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideE] = problem_size;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            if constexpr(std::is_same_v<decltype(layout), ck::tensor_layout::gemm::RowMajor>)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<D0DataType> d0_m_n(f_host_tensor_descriptor(M, N, StrideD0, D0Layout{}));
+    Tensor<D1DataType> d1_m_n(f_host_tensor_descriptor(M, N, StrideD1, D1Layout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d0_m_n: " << d0_m_n.mDesc << std::endl;
+    std::cout << "d1_m_n: " << d1_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(config.init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d0_m_n.GenerateTensorValue(GeneratorTensor_2<D0DataType>{-5, 5});
+        d1_m_n.GenerateTensorValue(GeneratorTensor_2<D1DataType>{-1, 1});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d0_m_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
+        d1_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d0_device_buf(sizeof(D0DataType) * d0_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem d1_device_buf(sizeof(D1DataType) * d1_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d0_device_buf.ToDevice(d0_m_n.mData.data());
+    d1_device_buf.ToDevice(d1_m_n.mData.data());
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                               b_device_buf.GetDeviceBuffer(),
+                               {d0_device_buf.GetDeviceBuffer(), d1_device_buf.GetDeviceBuffer()},
+                               e_device_buf.GetDeviceBuffer(),
+                               M,
+                               N,
+                               K,
+                               StrideA,
+                               StrideB,
+                               {StrideD0, StrideD1},
+                               StrideE,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        std::cout << "wrong! this device_op instance does not support this problem" << std::endl;
+        return true;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+
+    std::size_t flop      = 2_uz * M * N * K;
+    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                            sizeof(D0DataType) * N + sizeof(D1DataType) * M * N +
+                            sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << device_op.GetTypeString() << std::endl;
+
+    if(config.do_verification)
+    {
+        Tensor<AccDataType> c_m_n({M, N});
+
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d0_m_n(m, n), d1_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result);
+    }
+
+    return true;
+}
+
+bool run_gemm_add_multiply_example(int argc, char* argv[])
+{
+    ProblemSize problem_size;
+    ExecutionConfig config;
+
+    return !parse_cmd_args(argc, argv, problem_size, config) ||
+           run_gemm_add_multiply(problem_size, config);
+}
diff --git a/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt b/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..14432f6e23d4adbf09e0df6c805a68f1a5571f69
--- /dev/null
+++ b/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
@@ -0,0 +1,8 @@
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_example_executable(example_gemm_bias_softmax_gemm_permute gemm_bias_softmax_gemm_permute.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp b/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a90a6340a431a55c271dca3d3d0d1771382218f5
--- /dev/null
+++ b/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
@@ -0,0 +1,409 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using AElementOp    = ck::tensor_operation::element_wise::PassThrough;
+using B0ElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using C0DEElementOp = ck::tensor_operation::element_wise::ScaleAdd;
+using Acc0ElementOp = ck::tensor_operation::element_wise::PassThrough;
+using B1ElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CElementOp    = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKOPadding;
+constexpr static auto MaskingSpec =
+    ck::tensor_operation::device::MaskingSpecialization::MaskDisabled;
+static constexpr auto TensorSpecA  = ck::tensor_operation::device::TensorSpecialization::Default;
+static constexpr auto TensorSpecB0 = ck::tensor_operation::device::TensorSpecialization::Default;
+static constexpr auto TensorSpecB1 = ck::tensor_operation::device::TensorSpecialization::Default;
+static constexpr auto TensorSpecC  = ck::tensor_operation::device::TensorSpecialization::Default;
+
+using F16              = ck::half_t;
+using F32              = float;
+using ADataType        = F16;
+using B0DataType       = F16;
+using B1DataType       = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using CDataType        = F16;
+using D0DataType       = F16;
+using Acc0BiasDataType = ck::Tuple<D0DataType>;
+using Acc1BiasDataType = ck::Tuple<>;
+
+static constexpr ck::index_t NumDimG = 2;
+static constexpr ck::index_t NumDimM = 1;
+static constexpr ck::index_t NumDimN = 1;
+static constexpr ck::index_t NumDimK = 1;
+static constexpr ck::index_t NumDimO = 1;
+
+using DeviceOpInstance =
+    ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<
+        NumDimG,
+        NumDimM,
+        NumDimN,
+        NumDimK,
+        NumDimO,
+        ADataType,
+        B0DataType,
+        B1DataType,
+        CDataType,
+        Acc0BiasDataType,
+        Acc1BiasDataType,
+        AccDataType,
+        CShuffleDataType,
+        AElementOp,
+        B0ElementOp,
+        C0DEElementOp,
+        B1ElementOp,
+        CElementOp,
+        GemmSpec,
+        TensorSpecA,
+        TensorSpecB0,
+        TensorSpecB1,
+        TensorSpecC,
+        1,
+        256,
+        128,         // MPerBlock
+        128,         // NPerBlock
+        32,          // KPerBlock
+        64,          // Gemm1NPerBlock
+        32,          // Gemm1KPerBlock
+        8,           // AK1
+        8,           // BK1
+        2,           // B1K1
+        32,          // MPerXDL
+        32,          // NPerXDL
+        1,           // MXdlPerWave
+        4,           // NXdlPerWave
+        2,           // Gemm1NXdlPerWave
+        S<4, 64, 1>, // ABlockTransfer
+        S<1, 0, 2>,
+        S<1, 0, 2>,
+        2,
+        8,
+        8,
+        true,
+        S<4, 64, 1>, // BBlockTransfer
+        S<1, 0, 2>,
+        S<1, 0, 2>,
+        2,
+        8,
+        8,
+        true,
+        S<16, 16, 1>, // B1BlockTransfer
+        S<0, 2, 1>,
+        S<0, 2, 1>,
+        1,
+        4,
+        2,
+        false,
+        1,              // CShuffleMXdlPerWavePerShuffle
+        2,              // CShuffleNXdlPerWavePerShuffle
+        S<1, 32, 1, 8>, // CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
+        8,              // CShuffleBlockTransferScalarPerVector_NPerBlock
+        MaskingSpec,    // MaskingSpecialization
+        1>;
+
+// Ref Gemm0: fp16 in, fp32 out
+using ReferenceGemm0Instance = ck::tensor_operation::host::ReferenceBatchedGemm<ADataType,
+                                                                                B0DataType,
+                                                                                AccDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                B0ElementOp,
+                                                                                Acc0ElementOp>;
+
+// Ref Softmax: fp32 in, fp16 out
+using ReferenceSoftmaxInstance =
+    ck::tensor_operation::host::ReferenceSoftmax<AccDataType, ADataType, AccDataType>;
+
+// Ref Gemm1: fp16 in, fp16 out
+using ReferenceGemm1Instance = ck::tensor_operation::host::ReferenceBatchedGemm<ADataType,
+                                                                                B1DataType,
+                                                                                CDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                B1ElementOp,
+                                                                                CElementOp>;
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    int G0      = 3;
+    int G1      = 2;
+    int M       = 1024;
+    int N       = 1024;
+    int K       = 64;
+    int O       = 64;
+    float alpha = 1;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 11)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M  = std::stoi(argv[4]);
+        N  = std::stoi(argv[5]);
+        K  = std::stoi(argv[6]);
+        O  = std::stoi(argv[7]);
+        G0 = std::stoi(argv[8]);
+        G1 = std::stoi(argv[9]);
+
+        alpha = std::stof(argv[10]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 11: M, N, K, O, G0, G1\n");
+        printf("arg10: scale (alpha)\n");
+        exit(0);
+    }
+
+    std::vector<ck::index_t> a_gs_ms_ks_lengths{G0, G1, M, K};
+    std::vector<ck::index_t> a_gs_ms_ks_strides{
+        M * G1 * K, K, G1 * K, 1}; // A layout [G0, M, G1, K]
+
+    std::vector<ck::index_t> b0_gs_ns_ks_lengths{G0, G1, N, K};
+    std::vector<ck::index_t> b0_gs_ns_ks_strides{
+        N * G1 * K, K, G1 * K, 1}; // B0 layout [G0, N, G1, K]
+
+    std::vector<ck::index_t> b1_gs_os_ns_lengths{G0, G1, O, N};
+    std::vector<ck::index_t> b1_gs_os_ns_strides{
+        N * G1 * O, O, 1, G1 * O}; // B1 layout [G0, N, G1, O]
+
+    std::vector<ck::index_t> c_gs_ms_os_lengths{G0, G1, M, O};
+    std::vector<ck::index_t> c_gs_ms_os_strides{
+        M * G1 * O, O, G1 * O, 1}; // C layout [G0, M, G1, O]
+
+    // D layout [G0, M, G1, N]
+    std::vector<ck::index_t> d0_gs_ms_ns_lengths{G0, G1, M, N};
+    std::vector<ck::index_t> d0_gs_ms_ns_strides{M * G1 * N, N, G1 * N, 1};
+
+    Tensor<ADataType> a_gs_ms_ks(a_gs_ms_ks_lengths, a_gs_ms_ks_strides);
+    Tensor<B0DataType> b0_gs_ns_ks(b0_gs_ns_ks_lengths, b0_gs_ns_ks_strides);
+    Tensor<B1DataType> b1_gs_os_ns(b1_gs_os_ns_lengths, b1_gs_os_ns_strides);
+    Tensor<D0DataType> d0_gs_ms_ns(d0_gs_ms_ns_lengths, d0_gs_ms_ns_strides);
+    Tensor<CDataType> c_gs_ms_os_host_result(c_gs_ms_os_lengths, c_gs_ms_os_strides);
+    Tensor<CDataType> c_gs_ms_os_device_result(c_gs_ms_os_lengths, c_gs_ms_os_strides);
+
+    std::cout << "a_gs_ms_ks: " << a_gs_ms_ks.mDesc << std::endl;
+    std::cout << "b0_gs_ns_ks: " << b0_gs_ns_ks.mDesc << std::endl;
+    std::cout << "b1_gs_os_ns: " << b1_gs_os_ns.mDesc << std::endl;
+    std::cout << "c_gs_ms_os: " << c_gs_ms_os_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-2, 2});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_2<D0DataType>{-2, 2});
+        break;
+    case 2:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_3<B0DataType>{0.0, 1.0});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_3<B1DataType>{-0.5, 0.5});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_2<D0DataType>{-1, 1});
+        break;
+    case 3:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Diagonal<B0DataType>{});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_1<D0DataType>{1});
+        break;
+    default:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_Sequential<2>{});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Diagonal<B0DataType>{});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_1<D0DataType>{1});
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * G0 * G1 * M * K);
+    DeviceMem b0_device_buf(sizeof(B0DataType) * G0 * G1 * N * K);
+    DeviceMem d0_device_buf(sizeof(D0DataType) * G0 * G1 * M * N);
+    DeviceMem b1_device_buf(sizeof(B1DataType) * G0 * G1 * O * N);
+    DeviceMem c_device_buf(sizeof(CDataType) * G0 * G1 * M * O);
+
+    a_device_buf.ToDevice(a_gs_ms_ks.mData.data());
+    b0_device_buf.ToDevice(b0_gs_ns_ks.mData.data());
+    b1_device_buf.ToDevice(b1_gs_os_ns.mData.data());
+    d0_device_buf.ToDevice(d0_gs_ms_ns.mData.data());
+
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+
+    auto a_element_op    = AElementOp{};
+    auto b0_element_op   = B0ElementOp{};
+    auto c0de_element_op = C0DEElementOp{alpha};
+    auto acc0_element_op = Acc0ElementOp{};
+    auto b1_element_op   = B1ElementOp{};
+    auto c_element_op    = CElementOp{};
+
+    auto argument = device_op.MakeArgument(
+        static_cast<const ADataType*>(a_device_buf.GetDeviceBuffer()),
+        static_cast<const B0DataType*>(b0_device_buf.GetDeviceBuffer()),
+        static_cast<const B1DataType*>(b1_device_buf.GetDeviceBuffer()),
+        static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+        std::array<void*, 1>{d0_device_buf.GetDeviceBuffer()}, // p_acc0_biases
+        {},                                                    // p_acc1_biases
+        a_gs_ms_ks_lengths,
+        a_gs_ms_ks_strides,
+        b0_gs_ns_ks_lengths,
+        b0_gs_ns_ks_strides,
+        b1_gs_os_ns_lengths,
+        b1_gs_os_ns_strides,
+        c_gs_ms_os_lengths,
+        c_gs_ms_os_strides,
+        std::array<std::vector<ck::index_t>, 1>{
+            d0_gs_ms_ns_lengths}, // acc0_biases_gs_ms_ns_lengths
+        std::array<std::vector<ck::index_t>, 1>{
+            d0_gs_ms_ns_strides}, // acc0_biases_gs_ms_ns_strides
+        {},                       // acc1_biases_gs_ms_os_lengths
+        {},                       // acc1_biases_gs_ms_os_strides
+        a_element_op,
+        b0_element_op,
+        c0de_element_op,
+        b1_element_op,
+        c_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error("wrong! this device_op instance does not support this problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t BatchCount = G0 * G1;
+    std::size_t flop       = (size_t(M) * N * K * 2 + size_t(M) * N * O * 2) * BatchCount;
+    std::size_t num_btype =
+        (sizeof(ADataType) * M * K + sizeof(B0DataType) * K * N + sizeof(B1DataType) * N * O +
+         sizeof(CDataType) * M * O + sizeof(D0DataType) * M * N) *
+        BatchCount;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << std::endl;
+
+    if(do_verification)
+    {
+        c_device_buf.FromDevice(c_gs_ms_os_device_result.mData.data());
+
+        Tensor<ADataType> a_g_m_k({BatchCount, M, K});
+        Tensor<B0DataType> b0_g_k_n({BatchCount, K, N});
+        Tensor<B1DataType> b1_g_n_o({BatchCount, N, O});
+        Tensor<AccDataType> acc0_g_m_n({BatchCount, M, N});        // scratch object after gemm0
+        Tensor<ADataType> a1_g_m_n({BatchCount, M, N});            // scratch object after softmax
+        Tensor<CDataType> c_g_m_o_host_result({BatchCount, M, O}); // scratch object after gemm1
+        Tensor<D0DataType> d0_g_m_n({BatchCount, M, N});
+
+        // permute
+        a_gs_ms_ks.ForEach([&](auto& self, auto idx) {
+            a_g_m_k(idx[0] * G1 + idx[1], idx[2], idx[3]) = self(idx);
+        });
+        b0_gs_ns_ks.ForEach([&](auto& self, auto idx) {
+            b0_g_k_n(idx[0] * G1 + idx[1], idx[3], idx[2]) = self(idx);
+        });
+        b1_gs_os_ns.ForEach([&](auto& self, auto idx) {
+            b1_g_n_o(idx[0] * G1 + idx[1], idx[3], idx[2]) = self(idx);
+        });
+        d0_gs_ms_ns.ForEach([&](auto& self, auto idx) {
+            d0_g_m_n(idx[0] * G1 + idx[1], idx[2], idx[3]) = self(idx);
+        });
+
+        // gemm 0
+        auto ref_gemm0          = ReferenceGemm0Instance{};
+        auto ref_gemm0_invoker  = ref_gemm0.MakeInvoker();
+        auto ref_gemm0_argument = ref_gemm0.MakeArgument(
+            a_g_m_k, b0_g_k_n, acc0_g_m_n, a_element_op, b0_element_op, acc0_element_op);
+
+        ref_gemm0_invoker.Run(ref_gemm0_argument);
+
+        acc0_g_m_n.ForEach([&](auto&, auto idx) {
+            c0de_element_op(acc0_g_m_n(idx), acc0_g_m_n(idx), d0_g_m_n(idx));
+        });
+        // masking
+        const auto mask = DeviceOpInstance::C0MatrixMask(N);
+        acc0_g_m_n.ForEach([&](auto& self, auto idx) {
+            if(mask.IsMaskedElement(idx[1], idx[2]))
+                self(idx) = -ck::NumericLimits<float>::Infinity();
+        });
+
+        // softmax
+        auto ref_softmax          = ReferenceSoftmaxInstance{};
+        auto ref_softmax_invoker  = ref_softmax.MakeInvoker();
+        auto ref_softmax_argument = ref_softmax.MakeArgument(acc0_g_m_n, a1_g_m_n, 1, 0, {2});
+
+        ref_softmax_invoker.Run(ref_softmax_argument);
+
+        // gemm1
+        auto ref_gemm1          = ReferenceGemm1Instance{};
+        auto ref_gemm1_invoker  = ref_gemm1.MakeInvoker();
+        auto ref_gemm1_argument = ref_gemm1.MakeArgument(
+            a1_g_m_n, b1_g_n_o, c_g_m_o_host_result, PassThrough{}, b1_element_op, c_element_op);
+
+        ref_gemm1_invoker.Run(ref_gemm1_argument);
+
+        // permute
+        c_gs_ms_os_host_result.ForEach([&](auto& self, auto idx) {
+            const size_t& g0 = idx[0];
+            const size_t& g1 = idx[1];
+
+            const size_t g = g0 * G1 + g1;
+
+            self(idx) = c_g_m_o_host_result(g, idx[2], idx[3]);
+        });
+
+        // default absolute error and relative error is 0.001
+        double rtol = 1e-3;
+        double atol = 1e-3;
+
+        return ck::utils::check_err(c_gs_ms_os_device_result.mData,
+                                    c_gs_ms_os_host_result.mData,
+                                    "Error: Incorrect results!",
+                                    rtol,
+                                    atol)
+                   ? 0
+                   : 1;
+    }
+
+    return 0;
+}
diff --git a/example/48_pool3d_fwd/CMakeLists.txt b/example/48_pool3d_fwd/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5d58d6a0b16fd3b5482c02ff6b5ddd84a1e726a1
--- /dev/null
+++ b/example/48_pool3d_fwd/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
+
diff --git a/example/48_pool3d_fwd/pool3d_fwd_common.hpp b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..565bb94e47e13af73b4d49361f319c50252e84f0
--- /dev/null
+++ b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
@@ -0,0 +1,187 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+#pragma once
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/utility/reduction_enums.hpp"
+#include "ck/utility/reduction_functions_accumulate.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+
+template <typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          typename InLayout,
+          typename OutLayout,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+bool pool3d_test(bool do_verification,
+                 bool time_kernel,
+                 ck::index_t N,
+                 ck::index_t C,
+                 ck::index_t Z,
+                 ck::index_t Y,
+                 ck::index_t X,
+                 ck::index_t Di,
+                 ck::index_t Hi,
+                 ck::index_t Wi,
+                 ck::index_t window_stride_d,
+                 ck::index_t window_stride_h,
+                 ck::index_t window_stride_w,
+                 ck::index_t in_left_pad_d,
+                 ck::index_t in_left_pad_h,
+                 ck::index_t in_left_pad_w,
+                 ck::index_t in_right_pad_d,
+                 ck::index_t in_right_pad_h,
+                 ck::index_t in_right_pad_w)
+{
+    using DevicePoolFwdInstance =
+        ck::tensor_operation::device::DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<
+            InDataType,      // InDataType
+            OutDataType,     // OutDataType
+            IndexDataType,   // IndexDataType
+            ComputeDataType, // ComputeDataType
+            ReduceOpId,
+            OutputIndex,
+            64, // BlockSize
+            64, // ReduceMThreadClusterSize
+            1,  // ReduceKThreadClusterSize
+            4,  // ReduceMThreadSliceSize
+            1,  // ReduceKThreadSliceSize
+            4>; // InSrcOutDstVectorSize
+
+    const ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Z) / window_stride_d + 1;
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    const std::vector<ck::index_t> window_spatial_lengths{Z, Y, X};
+    const std::vector<ck::index_t> window_strides{
+        window_stride_d, window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> input_left_pads{in_left_pad_d, in_left_pad_h, in_left_pad_w};
+    const std::vector<ck::index_t> input_right_pads{in_right_pad_d, in_right_pad_h, in_right_pad_w};
+
+    // tensor layout
+    auto f_host_tensor_descriptor = [](std::size_t N_,
+                                       std::size_t C_,
+                                       std::size_t D,
+                                       std::size_t H,
+                                       std::size_t W,
+                                       auto layout) {
+        using namespace ck::literals;
+
+        if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+        {
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {C_ * D * H * W, D * H * W, H * W, W, 1_uz});
+        }
+        else if constexpr(ck::is_same<decltype(layout),
+                                      ck::tensor_layout::convolution::NDHWC>::value)
+        {
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+        }
+    };
+
+    Tensor<InDataType> in_n_c_di_hi_wi(f_host_tensor_descriptor(N, C, Di, Hi, Wi, InLayout{}));
+    Tensor<OutDataType> out_n_c_do_ho_wo_host(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_host(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+    Tensor<OutDataType> out_n_c_do_ho_wo_device(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_device(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+
+    std::cout << "in_n_c_di_hi_wi: " << in_n_c_di_hi_wi.mDesc << std::endl;
+    std::cout << "out_n_c_do_ho_wo: " << out_n_c_do_ho_wo_host.mDesc << std::endl;
+
+    in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_di_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
+                                     out_indices_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_di_hi_wi.mData.data());
+
+    auto pool         = DevicePoolFwdInstance{};
+    auto invoker_ptr  = pool.MakeInvokerPointer();
+    auto argument_ptr = pool.MakeArgumentPointer(
+        static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+        static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+        {N, C, Di, Hi, Wi},
+        {Z, Y, X},
+        {N, C, Do, Ho, Wo},
+        {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C},
+        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+        window_strides,
+        input_left_pads,
+        input_right_pads,
+        {2, 3, 4});
+
+    if(!pool.IsSupportedArgument(argument_ptr.get()))
+    {
+        throw std::runtime_error("wrong! device_op with the specified compilation parameters does "
+                                 "not support this problem");
+    }
+
+    float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+    std::cout << "Perf: " << ave_time << std::endl;
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+        using ReferencePoolingFwdInstance =
+            ck::tensor_operation::host::ReferencePoolingFwd<5,
+                                                            3,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ReduceOpId,
+                                                            PropagateNan,
+                                                            OutputIndex>;
+
+        auto ref_pooling          = ReferencePoolingFwdInstance{};
+        auto ref_pooling_invoker  = ref_pooling.MakeInvoker();
+        auto ref_pooling_argument = ref_pooling.MakeArgument(in_n_c_di_hi_wi,
+                                                             out_n_c_do_ho_wo_host,
+                                                             out_indices_n_c_do_ho_wo_host,
+                                                             window_spatial_lengths,
+                                                             window_strides,
+                                                             input_left_pads,
+                                                             input_right_pads);
+
+        ref_pooling_invoker.Run(ref_pooling_argument);
+
+        out_device_buf.FromDevice(out_n_c_do_ho_wo_device.mData.data());
+
+        pass = pass && ck::utils::check_err(out_n_c_do_ho_wo_device, out_n_c_do_ho_wo_host);
+
+        if constexpr(OutputIndex)
+        {
+            out_indices_device_buf.FromDevice(out_indices_n_c_do_ho_wo_device.mData.data());
+
+            pass = pass && ck::utils::check_err(out_indices_n_c_do_ho_wo_device,
+                                                out_indices_n_c_do_ho_wo_host);
+        };
+    }
+
+    return (pass);
+};
diff --git a/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..9afb51201deebe47b71932ab8fe1b9ffb50edcd6
--- /dev/null
+++ b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "pool3d_fwd_common.hpp"
+
+using InDataType      = ck::half_t;
+using OutDataType     = ck::half_t;
+using ComputeDataType = float;
+
+using IndexDataType = int32_t;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWC;
+using OutLayout = ck::tensor_layout::convolution::NDHWC;
+
+#if 1
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+#else
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+#endif
+
+static constexpr bool OutputIndex  = false;
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 2;
+    ck::index_t C               = 32;
+    ck::index_t Z               = 2;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Di              = 30;
+    ck::index_t Hi              = 30;
+    ck::index_t Wi              = 30;
+    ck::index_t window_stride_d = 2;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_d   = 1;
+    ck::index_t in_left_pad_h   = 1;
+    ck::index_t in_left_pad_w   = 1;
+    ck::index_t in_right_pad_d  = 1;
+    ck::index_t in_right_pad_h  = 1;
+    ck::index_t in_right_pad_w  = 1;
+
+    bool pass = pool3d_test<InDataType,
+                            OutDataType,
+                            ComputeDataType,
+                            IndexDataType,
+                            InLayout,
+                            OutLayout,
+                            ReduceOpId,
+                            PropagateNan,
+                            OutputIndex>(do_verification,
+                                         time_kernel,
+                                         N,
+                                         C,
+                                         Z,
+                                         Y,
+                                         X,
+                                         Di,
+                                         Hi,
+                                         Wi,
+                                         window_stride_d,
+                                         window_stride_h,
+                                         window_stride_w,
+                                         in_left_pad_d,
+                                         in_left_pad_h,
+                                         in_left_pad_w,
+                                         in_right_pad_d,
+                                         in_right_pad_h,
+                                         in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/49_maxpool2d_bwd/CMakeLists.txt b/example/49_maxpool2d_bwd/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b29cf9ccbce89d48bcff009d46b94e8c8e036741
--- /dev/null
+++ b/example/49_maxpool2d_bwd/CMakeLists.txt
@@ -0,0 +1,3 @@
+add_example_executable(example_maxpool2d_bwd_bf16 maxpool2d_bwd_bf16.cpp)
+add_example_executable(example_maxpool2d_bwd_fp16 maxpool2d_bwd_fp16.cpp)
+add_example_executable(example_maxpool2d_bwd_fp32 maxpool2d_bwd_fp32.cpp)
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..08a8009c6c35cc3f9c7881efb28daa571c96b5cd
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "maxpool2d_bwd_common.hpp"
+
+using InDataType      = ck::bhalf_t;
+using OutDataType     = ck::bhalf_t;
+using IndexDataType   = int32_t;
+using ComputeDataType = float;
+using DInDataType     = ck::bhalf_t;
+using DOutDataType    = ck::bhalf_t;
+
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 1;
+    ck::index_t C               = 1;
+    ck::index_t Y               = 3;
+    ck::index_t X               = 3;
+    ck::index_t Hi              = 32;
+    ck::index_t Wi              = 32;
+    ck::index_t window_stride_h = 1;
+    ck::index_t window_stride_w = 1;
+    ck::index_t in_left_pad_h   = 0;
+    ck::index_t in_left_pad_w   = 0;
+    ck::index_t in_right_pad_h  = 0;
+    ck::index_t in_right_pad_w  = 0;
+
+    bool pass = maxpool_bwd_test<InDataType,
+                                 OutDataType,
+                                 IndexDataType,
+                                 ComputeDataType,
+                                 DInDataType,
+                                 DOutDataType,
+                                 PropagateNan>(do_verification,
+                                               time_kernel,
+                                               N,
+                                               C,
+                                               Y,
+                                               X,
+                                               Hi,
+                                               Wi,
+                                               window_stride_h,
+                                               window_stride_w,
+                                               in_left_pad_h,
+                                               in_left_pad_w,
+                                               in_right_pad_h,
+                                               in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..045793cc2ae1ca9684ac5142821782d2c03cdb44
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
@@ -0,0 +1,222 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/utility/reduction_enums.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp"
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename ComputeDataType,
+          typename DInDataType,
+          typename DOutDataType,
+          bool PropagateNan>
+bool maxpool_bwd_test(bool do_verification,
+                      bool time_kernel,
+                      ck::index_t N,
+                      ck::index_t C,
+                      ck::index_t Y,
+                      ck::index_t X,
+                      ck::index_t Hi,
+                      ck::index_t Wi,
+                      ck::index_t window_stride_h,
+                      ck::index_t window_stride_w,
+                      ck::index_t in_left_pad_h,
+                      ck::index_t in_left_pad_w,
+                      ck::index_t in_right_pad_h,
+                      ck::index_t in_right_pad_w)
+{
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    using DevicePoolFwdInstance =
+        ck::tensor_operation::device::DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<
+            InDataType,      // InDataType
+            OutDataType,     // OutDataType
+            IndexDataType,   // IndexDataType
+            ComputeDataType, // ComputeDataType
+            ck::ReduceTensorOp::MAX,
+            true, // OutputIndex
+            64,   // BlockSize
+            64,   // ReduceMThreadClusterSize
+            1,    // ReduceKThreadClusterSize
+            4,    // ReduceMThreadSliceSize
+            1,    // ReduceKThreadSliceSize
+            1>;   // InSrcOutDstVectorSize
+
+    using DeviceMaxPoolBwdInstance = ck::tensor_operation::device::
+        DeviceIndexPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 4>;
+
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    const std::vector<ck::index_t> window_spatial_lengths{Y, X};
+    const std::vector<ck::index_t> window_strides{window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> input_left_pads{in_left_pad_h, in_left_pad_w};
+    const std::vector<ck::index_t> input_right_pads{in_right_pad_h, in_right_pad_w};
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+            // reference need Tensor with NCHW order
+            return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
+        };
+
+    // in
+    Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi));
+
+    // out
+    Tensor<OutDataType> out_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<OutDataType> out_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    // indices
+    Tensor<IndexDataType> indices_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<IndexDataType> indices_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    // dout
+    Tensor<DOutDataType> dout_n_c_ho_wo(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    // din
+    Tensor<DInDataType> din_n_c_hi_wi_host(f_host_tensor_descriptor(N, C, Hi, Wi));
+    Tensor<DInDataType> din_n_c_hi_wi_device(f_host_tensor_descriptor(N, C, Hi, Wi));
+
+    std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi.mDesc << std::endl;
+    std::cout << "out_n_c_ho_wo: " << out_n_c_ho_wo_host.mDesc << std::endl;
+    std::cout << "indices_n_c_ho_wo: " << indices_n_c_ho_wo_host.mDesc << std::endl;
+    std::cout << "dout_n_c_ho_wo: " << dout_n_c_ho_wo.mDesc << std::endl;
+    std::cout << "din_n_c_hi_wi: " << din_n_c_hi_wi_host.mDesc << std::endl;
+
+    in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
+    dout_n_c_ho_wo.GenerateTensorValue(GeneratorTensor_3<DOutDataType>{-1.0, 1.0});
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem indices_device_buf(sizeof(IndexDataType) *
+                                 indices_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem dout_device_buf(sizeof(DOutDataType) * dout_n_c_ho_wo.mDesc.GetElementSpaceSize());
+    DeviceMem din_device_buf(sizeof(DInDataType) *
+                             din_n_c_hi_wi_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
+    dout_device_buf.ToDevice(dout_n_c_ho_wo.mData.data());
+
+    auto pool_fwd              = DevicePoolFwdInstance{};
+    auto pool_fwd_invoker_ptr  = pool_fwd.MakeInvokerPointer();
+    auto pool_fwd_argument_ptr = pool_fwd.MakeArgumentPointer(
+        static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+        static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+        {N, C, Hi, Wi},
+        window_spatial_lengths,
+        {N, C, Ho, Wo},
+        {C * Hi * Wi, 1, Wi * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
+        window_strides,
+        input_left_pads,
+        input_right_pads,
+        {2, 3});
+
+    if(!pool_fwd.IsSupportedArgument(pool_fwd_argument_ptr.get()))
+    {
+        throw std::runtime_error("wrong! pool_fwd with the specified compilation parameters does "
+                                 "not support this problem");
+    }
+
+    float ave_time_fwd =
+        pool_fwd_invoker_ptr->Run(pool_fwd_argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    auto pool_bwd              = DeviceMaxPoolBwdInstance{};
+    auto pool_bwd_invoker_ptr  = pool_bwd.MakeInvokerPointer();
+    auto pool_bwd_argument_ptr = pool_bwd.MakeArgumentPointer(
+        static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+        static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+        dout_n_c_ho_wo.mDesc.GetElementSpaceSize(),
+        din_n_c_hi_wi_device.mDesc.GetElementSpaceSize(),
+        window_spatial_lengths,
+        window_strides);
+
+    if(!pool_bwd.IsSupportedArgument(pool_bwd_argument_ptr.get()))
+    {
+        throw std::runtime_error("wrong! pool_bwd with the specified compilation parameters does "
+                                 "not support this problem");
+    }
+
+    size_t pool_bwd_workspace_sz = pool_bwd.GetWorkSpaceSize(pool_bwd_argument_ptr.get());
+    DeviceMem pool_bwd_workspace_device_buf(pool_bwd_workspace_sz);
+    pool_bwd.SetWorkSpacePointer(pool_bwd_argument_ptr.get(),
+                                 pool_bwd_workspace_device_buf.GetDeviceBuffer());
+
+    float ave_time_bwd =
+        pool_bwd_invoker_ptr->Run(pool_bwd_argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    std::cout << "Pool fwd perf: " << ave_time_fwd << " ms" << std::endl;
+    std::cout << "Pool bwd perf: " << ave_time_bwd << " ms" << std::endl;
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+        using ReferencePoolingFwdInstance =
+            ck::tensor_operation::host::ReferencePoolingFwd<4,
+                                                            2,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ck::ReduceTensorOp::MAX,
+                                                            PropagateNan,
+                                                            true>;
+
+        auto ref_pooling_fwd          = ReferencePoolingFwdInstance{};
+        auto ref_pooling_fwd_invoker  = ref_pooling_fwd.MakeInvoker();
+        auto ref_pooling_fwd_argument = ref_pooling_fwd.MakeArgument(in_n_c_hi_wi,
+                                                                     out_n_c_ho_wo_host,
+                                                                     indices_n_c_ho_wo_host,
+                                                                     window_spatial_lengths,
+                                                                     window_strides,
+                                                                     input_left_pads,
+                                                                     input_right_pads);
+        ref_pooling_fwd_invoker.Run(ref_pooling_fwd_argument);
+
+        using ReferencePoolingBwdInstance =
+            ck::tensor_operation::host::ReferenceMaxPoolBwd<DOutDataType,
+                                                            IndexDataType,
+                                                            ComputeDataType,
+                                                            DInDataType,
+                                                            PassThrough>;
+
+        auto ref_pooling_bwd          = ReferencePoolingBwdInstance{};
+        auto ref_pooling_bwd_invoker  = ref_pooling_bwd.MakeInvoker();
+        auto ref_pooling_bwd_argument = ref_pooling_bwd.MakeArgument(
+            dout_n_c_ho_wo, indices_n_c_ho_wo_host, din_n_c_hi_wi_host, PassThrough{});
+
+        ref_pooling_bwd_invoker.Run(ref_pooling_bwd_argument);
+
+        out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
+        indices_device_buf.FromDevice(indices_n_c_ho_wo_device.mData.data());
+        din_device_buf.FromDevice(din_n_c_hi_wi_device.mData.data());
+
+        pass = pass && ck::utils::check_err(out_n_c_ho_wo_device, out_n_c_ho_wo_host);
+        pass = pass && ck::utils::check_err(indices_n_c_ho_wo_device, indices_n_c_ho_wo_host);
+        pass = pass && ck::utils::check_err(din_n_c_hi_wi_device, din_n_c_hi_wi_host);
+    }
+
+    return (pass);
+};
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c6b0b2c86006ddb30788a06ffac7fdb177f3abca
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "maxpool2d_bwd_common.hpp"
+
+using InDataType      = ck::half_t;
+using OutDataType     = ck::half_t;
+using IndexDataType   = int32_t;
+using ComputeDataType = float;
+using DInDataType     = ck::half_t;
+using DOutDataType    = ck::half_t;
+
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 1;
+    ck::index_t C               = 1;
+    ck::index_t Y               = 3;
+    ck::index_t X               = 3;
+    ck::index_t Hi              = 32;
+    ck::index_t Wi              = 32;
+    ck::index_t window_stride_h = 1;
+    ck::index_t window_stride_w = 1;
+    ck::index_t in_left_pad_h   = 0;
+    ck::index_t in_left_pad_w   = 0;
+    ck::index_t in_right_pad_h  = 0;
+    ck::index_t in_right_pad_w  = 0;
+
+    bool pass = maxpool_bwd_test<InDataType,
+                                 OutDataType,
+                                 IndexDataType,
+                                 ComputeDataType,
+                                 DInDataType,
+                                 DOutDataType,
+                                 PropagateNan>(do_verification,
+                                               time_kernel,
+                                               N,
+                                               C,
+                                               Y,
+                                               X,
+                                               Hi,
+                                               Wi,
+                                               window_stride_h,
+                                               window_stride_w,
+                                               in_left_pad_h,
+                                               in_left_pad_w,
+                                               in_right_pad_h,
+                                               in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c79b84c489d637e3c3ec9b52fd726537af3f776b
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "maxpool2d_bwd_common.hpp"
+
+using InDataType      = float;
+using OutDataType     = float;
+using IndexDataType   = int32_t;
+using ComputeDataType = float;
+using DInDataType     = float;
+using DOutDataType    = float;
+
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 1;
+    ck::index_t C               = 1;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Hi              = 32;
+    ck::index_t Wi              = 32;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_h   = 0;
+    ck::index_t in_left_pad_w   = 0;
+    ck::index_t in_right_pad_h  = 0;
+    ck::index_t in_right_pad_w  = 0;
+
+    bool pass = maxpool_bwd_test<InDataType,
+                                 OutDataType,
+                                 IndexDataType,
+                                 ComputeDataType,
+                                 DInDataType,
+                                 DOutDataType,
+                                 PropagateNan>(do_verification,
+                                               time_kernel,
+                                               N,
+                                               C,
+                                               Y,
+                                               X,
+                                               Hi,
+                                               Wi,
+                                               window_stride_h,
+                                               window_stride_w,
+                                               in_left_pad_h,
+                                               in_left_pad_w,
+                                               in_right_pad_h,
+                                               in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/50_put_element/CMakeLists.txt b/example/50_put_element/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1b0020ebcf65a87ad39b004c149385cdd7de89ae
--- /dev/null
+++ b/example/50_put_element/CMakeLists.txt
@@ -0,0 +1 @@
+add_example_executable(example_put_element_fp16 put_element_fp16.cpp)
diff --git a/example/50_put_element/put_element_fp16.cpp b/example/50_put_element/put_element_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d4b6831bc4713bac25358998df3e9abfb9785e80
--- /dev/null
+++ b/example/50_put_element/put_element_fp16.cpp
@@ -0,0 +1,88 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+using XDataType     = ck::half_t;
+using YDataType     = ck::half_t;
+using IndexDataType = int32_t;
+
+using YElementwiseOp = ck::tensor_operation::element_wise::PassThrough;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DevicePutElementImpl<XDataType,     // XDataType
+                                                       IndexDataType, // IndexDataType
+                                                       YDataType,     // YDataType
+                                                       YElementwiseOp,
+                                                       ck::InMemoryDataOperationEnum::Set,
+                                                       1>;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    int N = 1024;
+
+    Tensor<XDataType> x(HostTensorDescriptor{N, 1});
+    Tensor<IndexDataType> indices(HostTensorDescriptor{N, 1});
+    Tensor<YDataType> y(HostTensorDescriptor{N, 1});
+
+    x.GenerateTensorValue(GeneratorTensor_3<XDataType>{-1.0, 1.0});
+    for(int i = 0; i < N; ++i)
+        indices(i) = i;
+
+    DeviceMem x_device_buf(sizeof(XDataType) * x.mDesc.GetElementSpaceSize());
+    DeviceMem y_device_buf(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
+    DeviceMem indices_device_buf(sizeof(IndexDataType) * indices.mDesc.GetElementSpaceSize());
+
+    x_device_buf.ToDevice(x.mData.data());
+    indices_device_buf.ToDevice(indices.mData.data());
+
+    auto put_instance     = DeviceInstance{};
+    auto put_invoker_ptr  = put_instance.MakeInvokerPointer();
+    auto put_argument_ptr = put_instance.MakeArgumentPointer(
+        static_cast<XDataType*>(x_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+        static_cast<YDataType*>(y_device_buf.GetDeviceBuffer()),
+        N,
+        N,
+        YElementwiseOp{});
+
+    if(!put_instance.IsSupportedArgument(put_argument_ptr.get()))
+    {
+        throw std::runtime_error("argument is not supported!");
+    }
+
+    float ave_time =
+        put_invoker_ptr->Run(put_argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    std::cout << "perf: " << ave_time << " ms" << std::endl;
+
+    bool pass = true;
+    if(do_verification)
+    {
+        Tensor<YDataType> y_host(HostTensorDescriptor{N, 1});
+
+        for(int i = 0; i < N; ++i)
+        {
+            IndexDataType idx = indices(i);
+            y_host(idx)       = x(i);
+        }
+
+        y_device_buf.FromDevice(y.mData.data());
+        pass = ck::utils::check_err(y, y_host);
+    }
+
+    return (pass ? 0 : 1);
+}
diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index c68bd6f50f76dfad341588ee02d482794ca1949e..6a1ac0902180d5889be2b73cb5e1be3ff4abbba6 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -18,36 +18,49 @@
 #define CK_USE_LAUNCH_BOUNDS 1
 
 #ifdef CK_USE_LAUNCH_BOUNDS
+// for most kernels
 #define CK_MAX_THREAD_PER_BLOCK 256
 #define CK_MIN_BLOCK_PER_CU 2
+
+// for wavelet GEMM kernel
+#define CK_WAVELET_MAX_THREAD_PER_BLOCK 512
+#define CK_WAVELET_MIN_BLOCK_PER_CU 2
+#endif
+
+// kernel attribute: amdgpu_waves_per_eu()
+#ifdef CK_USE_WAVES_PER_EU
+// for 1-wave kernels, control arguments of amdgpu_waves_per_eu() attribute
+#ifndef CK_MIN_WAVES_PER_EU
+#define CK_MIN_WAVES_PER_EU 0
 #endif
 
-// check GPU target
-#ifdef __HIP_DEVICE_COMPILE__
-#if !(defined(__gfx803__) || defined(__gfx900__) || defined(__gfx906__) || defined(__gfx908__) || \
-      defined(__gfx90a__) || defined(__gfx1030__) || defined(__gfx1100__))
-#error Not supported target
+#ifndef CK_MAX_WAVES_PER_EU
+#define CK_MAX_WAVES_PER_EU 0
 #endif
+
+#else
+#define CK_USE_WAVES_PER_EU 0
 #endif
 
 // buffer resource
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_BUFFER_RESOURCE_3RD_DWORD -1
 #elif defined(__gfx803__) || defined(__gfx900__) || defined(__gfx906__) || defined(__gfx908__) || \
-    defined(__gfx90a__) // for GPU code
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) ||                          \
+    defined(__gfx942__) // for GPU code
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x00020000
 #elif defined(__gfx1030__) // for GPU code
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
-#elif defined(__gfx1100__) // for GPU code
-#define CK_BUFFER_RESOURCE_3RD_DWORD 0x10020000
+#elif defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__) // for GPU code
+#define CK_BUFFER_RESOURCE_3RD_DWORD 0x31004000
 #endif
 
 // FMA instruction
 #ifndef __HIP_DEVICE_COMPILE__                   // for host code, define nothing
 #elif defined(__gfx803__) || defined(__gfx900__) // for GPU code
 #define CK_USE_AMD_V_MAC_F32
-#elif defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx1030__) // for GPU code
+#elif defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__) // for GPU code
 #define CK_USE_AMD_V_FMAC_F32
 #define CK_USE_AMD_V_DOT2_F32_F16
 #define CK_USE_AMD_V_DOT4_I32_I8
@@ -56,18 +69,23 @@
 // MFMA instruction
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_MFMA
-#elif defined(__gfx908__) || defined(__gfx90a__) // for GPU code
+#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
+    defined(__gfx942__) // for GPU code
 #define CK_USE_AMD_MFMA
 #endif
 
-#if defined(__gfx90a__)
+#if(defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
 #define CK_USE_AMD_MFMA_BF16_1K_OP
 #endif
 
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+#define CK_USE_AMD_MFMA_GFX940
+#endif
+
 // WMMA instruction
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_WMMA
-#elif defined(__gfx1100__) // for GPU code
+#elif defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__) // for GPU code
 #define CK_USE_AMD_WMMA
 #endif
 
@@ -83,13 +101,15 @@
 // buffer atomic add: floating point
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
-#elif defined(__gfx908__) || defined(__gfx90a__) // for GPU code
+#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
+    defined(__gfx942__) // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
 #else // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 0
 #endif
 
-#if defined(__gfx90a__) // for GPU code
+#if(defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
+    defined(__gfx942__)) // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 1
 #else
 #define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 0
@@ -143,6 +163,10 @@
 #define CK_EXPERIMENTAL_INTER_WAVE_INSTANCES 1
 // experimental feature: add instances using pipeline v2
 #define CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES 1
+// experimental feature: optimize pipeline v2 by IGLP strategy (value=ID of strategy)
+#ifndef CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT
+#define CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT 0
+#endif
 
 // hack: have underlying assumption that need to be satsified, otherwise it's a bug
 // hack for forcing register to keep idx_diff_low_const in SGPR. idx_diff_low_const must be
@@ -163,16 +187,26 @@
 // tuning parameter
 #define CK_WORKAROUND_SWDEV_325164 0
 
-// workaround: a BF16 attention kernel for gfx908 is likely affected by a compiler issue
-#ifdef __gfx908__
-#define CK_WORKAROUND_SWDEV_XXXXXX_BF16_ATTEN_FWD_GFX908_ISSUE 1
-#else // __gfx90a__, ...
-#define CK_WORKAROUND_SWDEV_XXXXXX_BF16_ATTEN_FWD_GFX908_ISSUE 0
-#endif // __gfx908__
+// workaround: compiler not emiting reciprocal instruction frm __frcp_rn()
+#define CK_WORKAROUND_SWDEV_383542 1
+
+// workaround: compiler issue on gfx908
+#define CK_WORKAROUND_SWDEV_388832 1
+
+// workaround: Grouped Conv2d_bwd_data fails for already implemented instance
+#define CK_WORKAROUND_SWDEV_3318619 0
 
 // flag to enable (1) or disable (0) the debugging output in some kernels
 #define DEBUG_LOG 0
 
+// denorm test fix, required to work around dissue
+#ifndef CK_WORKAROUND_DENORM_FIX
+#define CK_WORKAROUND_DENORM_FIX 0
+#elif
+// enable only on MI200
+#define CK_WORKAROUND_DENORM_FIX = CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
+#endif // CK_WORKAROUND_DENORM_FIX
+
 namespace ck {
 
 enum struct InMemoryDataOperationEnum
diff --git a/include/ck/host_utility/device_prop.hpp b/include/ck/host_utility/device_prop.hpp
index e2cbdb733272d37aa5dbc7a746e86911d6b8644f..bd02d5d88a2b0a6f882ac58bfada67649e210bc0 100644
--- a/include/ck/host_utility/device_prop.hpp
+++ b/include/ck/host_utility/device_prop.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/host_utility/hip_check_error.hpp b/include/ck/host_utility/hip_check_error.hpp
index d3dc8eaf1eb8b87207256ea4a521e23d5a49ca9c..af7bebd9d6afbf59f1852156b1708ed843d9a9f6 100644
--- a/include/ck/host_utility/hip_check_error.hpp
+++ b/include/ck/host_utility/hip_check_error.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/host_utility/io.hpp b/include/ck/host_utility/io.hpp
index ac8719592db8b48996bfbc65dc656d6d96bde545..55734bab2e469a12cd02c417cbb25d9c8a49727c 100644
--- a/include/ck/host_utility/io.hpp
+++ b/include/ck/host_utility/io.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/host_utility/kernel_launch.hpp b/include/ck/host_utility/kernel_launch.hpp
index ed6e2f0ba1d0ca433487fa32f374fde3b7ecee0e..58740b43517b9e75bdab59d2e8134d5c6dd97854 100644
--- a/include/ck/host_utility/kernel_launch.hpp
+++ b/include/ck/host_utility/kernel_launch.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -20,6 +20,7 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
 #if CK_TIME_KERNEL
     if(stream_config.time_kernel_)
     {
+#if DEBUG_LOG
         printf("%s: grid_dim {%d, %d, %d}, block_dim {%d, %d, %d} \n",
                __func__,
                grid_dim.x,
@@ -29,15 +30,15 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
                block_dim.y,
                block_dim.z);
 
-        const int nrepeat = 10;
-
         printf("Warm up 1 time\n");
-
+#endif
         // warm up
         kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
 
+        const int nrepeat = 10;
+#if DEBUG_LOG
         printf("Start running %d times...\n", nrepeat);
-
+#endif
         hipEvent_t start, stop;
 
         hip_check_error(hipEventCreate(&start));
diff --git a/include/ck/host_utility/stream_utility.hpp b/include/ck/host_utility/stream_utility.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..9ab49489bbe59b5152ad61c74622be48c33c28a7
--- /dev/null
+++ b/include/ck/host_utility/stream_utility.hpp
@@ -0,0 +1,43 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <hip/hip_runtime.h>
+
+#include "ck/stream_config.hpp"
+#include "ck/host_utility/hip_check_error.hpp"
+
+static inline int getAvailableComputeUnitCount(const StreamConfig& stream_config)
+{
+    constexpr int MAX_MASK_DWORDS = 64;
+
+    // assume at most 64*32 = 2048 CUs
+    uint32_t cuMask[MAX_MASK_DWORDS];
+
+    for(int i = 0; i < MAX_MASK_DWORDS; i++)
+        cuMask[i] = 0;
+
+    auto countSetBits = [](uint32_t dword) {
+        int count = 0;
+
+        while(dword != 0)
+        {
+            if(dword & 0x1)
+                count++;
+
+            dword = dword >> 1;
+        };
+
+        return (count);
+    };
+
+    hip_check_error(hipExtStreamGetCUMask(stream_config.stream_id_, MAX_MASK_DWORDS, &cuMask[0]));
+
+    int ret = 0;
+
+    for(int i = 0; i < MAX_MASK_DWORDS; i++)
+        ret += countSetBits(cuMask[i]);
+
+    return (ret);
+};
diff --git a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index db8e48df6d4b18e9c8a60b8f3d4382d1c6e649f8..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,275 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// Number of GEMMs = YTilde * XTilde
-// GemmM = C
-// GemmN = N * HTildeSlice * WTildeSlice
-// GemmK = K * YDotSlice * XDotSlice
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t IYTildeValue,
-          index_t IXTildeValue,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<IYTildeValue>,
-    Number<IXTildeValue>,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1  = Number<GemmK1Value>{};
-    constexpr auto IYTilde = Number<IYTildeValue>{};
-    constexpr auto IXTilde = Number<IXTildeValue>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
-    const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
-
-    const auto YTilde = ConvStrideH / GcdStrideDilationH;
-    const auto XTilde = ConvStrideW / GcdStrideDilationW;
-
-    const auto YDot = math::integer_divide_ceil(Y, YTilde);
-    const auto XDot = math::integer_divide_ceil(X, XTilde);
-
-    const auto HTilde = Ho + math::integer_divide_ceil(ConvDilationH * (Y - I1), ConvStrideH);
-    const auto WTilde = Wo + math::integer_divide_ceil(ConvDilationW * (X - I1), ConvStrideW);
-
-    // only work on HTilde and WTilde that contribute to non-padding area of input tensor
-    const auto IHTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadH - ConvDilationH * (YTilde - I1)), ConvStrideH);
-    const auto IWTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadW - ConvDilationW * (XTilde - I1)), ConvStrideW);
-
-    const auto IHTildeSliceEnd =
-        math::min(HTilde, math::integer_divide_ceil(InLeftPadH + Hi - I1, ConvStrideH) + I1);
-    const auto IWTildeSliceEnd =
-        math::min(WTilde, math::integer_divide_ceil(InLeftPadW + Wi - I1, ConvStrideW) + I1);
-
-    const auto HTildeSlice = IHTildeSliceEnd - IHTildeSliceBegin;
-    const auto WTildeSlice = IWTildeSliceEnd - IWTildeSliceBegin;
-
-    // GemmK is different for each GEMM
-    const auto YDotSlice = math::integer_divide_ceil(Y - IYTilde, YTilde);
-    const auto XDotSlice = math::integer_divide_ceil(X - IXTilde, XTilde);
-
-    const auto K1 = GemmK1;
-    const auto K0 = K / K1;
-
-    // weight tensor
-    const auto wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc = transform_tensor_descriptor(
-        wei_k_y_x_c_grid_desc,
-        make_tuple(make_pass_through_transform(K),
-                   make_embed_transform(make_tuple(YDot, YTilde),
-                                        make_tuple(ConvStrideH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, XTilde),
-                                        make_tuple(ConvStrideW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto wei_k0_k1_ydotslice_xdotslice_c_grid_desc =
-        transform_tensor_descriptor(wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                                               make_slice_transform(YDot, I0, YDotSlice),
-                                               make_slice_transform(XDot, I0, XDotSlice),
-                                               make_freeze_transform(IYTilde),
-                                               make_freeze_transform(IXTilde),
-                                               make_pass_through_transform(C)),
-                                    make_tuple(Sequence<0>{},
-                                               Sequence<1>{},
-                                               Sequence<3>{},
-                                               Sequence<2>{},
-                                               Sequence<4>{},
-                                               Sequence<5>{}),
-                                    make_tuple(Sequence<0, 1>{},
-                                               Sequence<2>{},
-                                               Sequence<3>{},
-                                               Sequence<>{},
-                                               Sequence<>{},
-                                               Sequence<4>{}));
-
-#if 1
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<2, 3, 0>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<0, 2, 3>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // output tensor
-    // this add padding check
-    const auto out_n_hop_wop_k_grid_desc = transform_tensor_descriptor(
-        out_n_ho_wo_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Ho, I0, I0),
-                   make_pad_transform(Wo, I0, I0),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto out_n_ydot_htilde_xdot_wtilde_k_grid_desc = transform_tensor_descriptor(
-        out_n_hop_wop_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YDot, HTilde),
-                                        make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, WTilde),
-                                        make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc =
-        transform_tensor_descriptor(
-            out_n_ydot_htilde_xdot_wtilde_k_grid_desc,
-            make_tuple(make_pass_through_transform(N),
-                       make_slice_transform(YDot, I0, YDotSlice),
-                       make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                       make_slice_transform(XDot, I0, XDotSlice),
-                       make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                       make_unmerge_transform(make_tuple(K0, K1))),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5>{}),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5, 6>{}));
-
-#if 1
-    const auto out_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto out_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<5, 1, 3>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YTilde, HTilde),
-                                        make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(XTilde, WTilde),
-                                        make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_n_htildeslice_wtildeslice_c_grid_desc = transform_tensor_descriptor(
-        in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_freeze_transform(IYTilde),
-                   make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                   make_freeze_transform(IXTilde),
-                   make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{},
-                   Sequence<1>{},
-                   Sequence<2>{},
-                   Sequence<3>{},
-                   Sequence<4>{},
-                   Sequence<5>{}),
-        make_tuple(Sequence<0>{},
-                   Sequence<>{},
-                   Sequence<1>{},
-                   Sequence<>{},
-                   Sequence<2>{},
-                   Sequence<3>{}));
-
-    const auto in_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        in_n_htildeslice_wtildeslice_c_grid_desc,
-        make_tuple(make_pass_through_transform(C),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice))),
-        make_tuple(Sequence<3>{}, Sequence<0, 1, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(wei_gemmk0_gemmm_gemmk1_grid_desc,
-                      out_gemmk0_gemmn_gemmk1_grid_desc,
-                      in_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 5391b595b5c1e9b5b7395d1a676efcd87b9f2e6e..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,355 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: out
-// B: wei
-// C: in
-// Number of GEMMs = YTilde * XTilde
-// GemmM = N * HTildeSlice * WTildeSlice
-// GemmN = C
-// GemmK = K * YDotSlice * XDotSlice
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          typename IYTilde,
-          typename IXTilde,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk(
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    IYTilde i_ytilde,
-    IXTilde i_xtilde,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
-    const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
-
-    const auto YTilde = ConvStrideH / GcdStrideDilationH;
-    const auto XTilde = ConvStrideW / GcdStrideDilationW;
-
-    const auto YDot = math::integer_divide_ceil(Y, YTilde);
-    const auto XDot = math::integer_divide_ceil(X, XTilde);
-
-    const auto HTilde = Ho + math::integer_divide_ceil(ConvDilationH * (Y - I1), ConvStrideH);
-    const auto WTilde = Wo + math::integer_divide_ceil(ConvDilationW * (X - I1), ConvStrideW);
-
-    // only work on HTilde and WTilde that contribute to non-padding area of input tensor
-    const auto IHTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadH - ConvDilationH * (YTilde - I1)), ConvStrideH);
-    const auto IWTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadW - ConvDilationW * (XTilde - I1)), ConvStrideW);
-
-    const auto IHTildeSliceEnd =
-        math::min(HTilde, math::integer_divide_ceil(InLeftPadH + Hi - I1, ConvStrideH) + I1);
-    const auto IWTildeSliceEnd =
-        math::min(WTilde, math::integer_divide_ceil(InLeftPadW + Wi - I1, ConvStrideW) + I1);
-
-    const auto HTildeSlice = IHTildeSliceEnd - IHTildeSliceBegin;
-    const auto WTildeSlice = IWTildeSliceEnd - IWTildeSliceBegin;
-
-    // GemmK is different for each GEMM
-    const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
-    const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
-
-    const auto K1 = GemmK1;
-    const auto K0 = K / K1;
-
-    // A: output tensor
-    // this add padding check
-    const auto out_n_hop_wop_k_grid_desc = transform_tensor_descriptor(
-        out_n_ho_wo_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Ho, I0, I0),
-                   make_pad_transform(Wo, I0, I0),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto out_n_ydot_htilde_xdot_wtilde_k_grid_desc = transform_tensor_descriptor(
-        out_n_hop_wop_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YDot, HTilde),
-                                        make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, WTilde),
-                                        make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc =
-        transform_tensor_descriptor(
-            out_n_ydot_htilde_xdot_wtilde_k_grid_desc,
-            make_tuple(make_pass_through_transform(N),
-                       make_slice_transform(YDot, I0, YDotSlice),
-                       make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                       make_slice_transform(XDot, I0, XDotSlice),
-                       make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                       make_unmerge_transform(make_tuple(K0, K1))),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5>{}),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5, 6>{}));
-
-#if 1
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<5, 1, 3>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // B: weight tensor
-    const auto wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc = transform_tensor_descriptor(
-        wei_k_y_x_c_grid_desc,
-        make_tuple(make_pass_through_transform(K),
-                   make_embed_transform(make_tuple(YDot, YTilde),
-                                        make_tuple(ConvStrideH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, XTilde),
-                                        make_tuple(ConvStrideW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto wei_k0_k1_ydotslice_xdotslice_c_grid_desc =
-        transform_tensor_descriptor(wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                                               make_slice_transform(YDot, I0, YDotSlice),
-                                               make_slice_transform(XDot, I0, XDotSlice),
-                                               make_freeze_transform(i_ytilde),
-                                               make_freeze_transform(i_xtilde),
-                                               make_pass_through_transform(C)),
-                                    make_tuple(Sequence<0>{},
-                                               Sequence<1>{},
-                                               Sequence<3>{},
-                                               Sequence<2>{},
-                                               Sequence<4>{},
-                                               Sequence<5>{}),
-                                    make_tuple(Sequence<0, 1>{},
-                                               Sequence<2>{},
-                                               Sequence<3>{},
-                                               Sequence<>{},
-                                               Sequence<>{},
-                                               Sequence<4>{}));
-
-#if 1
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<2, 3, 0>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<0, 2, 3>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // C: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YTilde, HTilde),
-                                        make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(XTilde, WTilde),
-                                        make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_n_htildeslice_wtildeslice_c_grid_desc = transform_tensor_descriptor(
-        in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_freeze_transform(i_ytilde),
-                   make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                   make_freeze_transform(i_xtilde),
-                   make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{},
-                   Sequence<1>{},
-                   Sequence<2>{},
-                   Sequence<3>{},
-                   Sequence<4>{},
-                   Sequence<5>{}),
-        make_tuple(Sequence<0>{},
-                   Sequence<>{},
-                   Sequence<1>{},
-                   Sequence<>{},
-                   Sequence<2>{},
-                   Sequence<3>{}));
-
-    const auto in_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        in_n_htildeslice_wtildeslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0, 1, 2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmk0_gemmm_gemmk1_grid_desc,
-                      wei_gemmk0_gemmn_gemmk1_grid_desc,
-                      in_gemmm_gemmn_grid_desc);
-}
-
-// A: out
-// B: wei
-// C: in
-// Number of GEMMs = 1
-// GemmM = N * Ho * Wo
-// GemmN = C
-// GemmK = K
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk_1x1(
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const TensorDescriptor<Wei...>& /* wei_k_y_x_c_grid_desc */,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const ConvStrides& conv_strides,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto K1 = GemmK1;
-    const auto K0 = K / K1;
-
-    // A: output tensor
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-                                    make_tuple(make_pass_through_transform(N * Ho * Wo),
-                                               make_unmerge_transform(make_tuple(K0, K1))),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0, 2>{}));
-
-    // B: weight tensor
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C)),
-        make_tuple(make_unmerge_transform(make_tuple(K0, K1)), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // C: input tensor
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(I1, Ho), make_tuple(I1, ConvStrideH)),
-                   make_embed_transform(make_tuple(I1, Wo), make_tuple(I1, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        in_n_y_ho_x_wo_c_grid_desc,
-        make_tuple(make_freeze_transform(I0),
-                   make_freeze_transform(I0),
-                   make_merge_transform(make_tuple(N, Ho, Wo)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<1>{}, Sequence<3>{}, Sequence<0, 2, 4>{}, Sequence<5>{}),
-        make_tuple(Sequence<>{}, Sequence<>{}, Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmk0_gemmm_gemmk1_grid_desc,
-                      wei_gemmk0_gemmn_gemmk1_grid_desc,
-                      in_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index bb1dc239f4d7e6a5abade2dda4a17a98ab7d1a47..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,150 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmK = N * Ho * Wo
-// GemmN = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value,
-          typename GemmKBatchType,
-          typename GemmKPadType>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>,
-    GemmKBatchType GemmKBatch,
-    GemmKPadType GemmKPad)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM      = K;
-    const auto GemmN      = C * Y * X;
-    const auto GemmKTotal = N * Ho * Wo;
-    const index_t GemmK0  = GemmKPad / (GemmKBatch * GemmK1);
-
-    // A: output tensor
-    const auto out_gemmktotal_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-        out_gemmktotal_gemmm_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_gemmkpad_gemmm_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // B: input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmktotal_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-        in_gemmktotal_gemmn_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        in_gemmkpad_gemmn_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index ca530934e49dcccea388fd8501e5bf046f1c94af..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,132 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmK = N * Ho * Wo
-// GemmN = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = K;
-    const auto GemmN  = C * Y * X;
-    const auto GemmK  = N * Ho * Wo;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(out_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index e960f90c4bb1974ddfaaf9faf6a12d75d1db2f4d..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,150 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: in
-// B: wei
-// C: out
-// GemmM = N * Ho * Wo
-// GemmN = K
-// GemmK = Y * X * C
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value,
-          typename GemmKBatchType,
-          typename GemmKPadType>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>,
-    GemmKBatchType GemmKBatch,
-    GemmKPadType GemmKPad)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM      = Y * X * C;
-    const auto GemmN      = K;
-    const auto GemmKTotal = N * Ho * Wo;
-    const index_t GemmK0  = GemmKPad / (GemmKBatch * GemmK1);
-
-    // A: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmktotal_gemmm_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-        in_gemmktotal_gemmm_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        in_gemmkpad_gemmm_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // B: output tensor
-    const auto out_gemmktotal_gemmn_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-
-    const auto out_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-        out_gemmktotal_gemmn_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_gemmkpad_gemmn_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(in_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                      out_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 052bab423db04740c88721026a3cceb3fa77c292..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,135 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: in
-// B: wei
-// C: out
-// GemmM = N * Ho * Wo
-// GemmN = K
-// GemmK = Y * X * C
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = Y * X * C;
-    const auto GemmN  = K;
-    const auto GemmK  = N * Ho * Wo;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // A: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmm_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // B: output tensor
-    const auto out_gemmk_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(out_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(in_gemmk0_gemmm_gemmk1_grid_desc,
-                      out_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index c301a9e0c67b11c42fbe2a0a99c4f764049a5222..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,147 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: out
-// B: in
-// C: wei
-// GemmM = K
-// GemmN = Y * X * C
-// GemmKTotal = N * Ho * Wo
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value,
-          typename GemmKBatchType,
-          typename GemmKPadType>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>,
-    GemmKBatchType GemmKBatch,
-    GemmKPadType GemmKPad)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM      = K;
-    const auto GemmN      = Y * X * C;
-    const auto GemmKTotal = N * Ho * Wo;
-    const index_t GemmK0  = GemmKPad / (GemmKBatch * GemmK1);
-
-    // A: output tensor
-    const auto out_gemmktotal_gemmm_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-
-    const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-        out_gemmktotal_gemmm_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_gemmkpad_gemmm_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // B: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmktotal_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-        in_gemmktotal_gemmn_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        in_gemmkpad_gemmn_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
-
-    return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp b/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp
index 41267536551ea298e111f6d93b8e6f3f8f9ed475..6b118e972e67897f2c9e0cad3a8f959760f82f6c 100644
--- a/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
 #define CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index 381f9ac9d6f5a9b000eb44839c7198e4dc9f5d9b..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,260 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_global_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_global_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_global_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_global_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_global_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_global_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_global_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_global_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_global_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_global_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_global_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_global_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_global_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_global_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_global_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_global_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_global_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_global_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_global_desc, in_gemmk_gemmn_global_desc, out_gemmm_gemmn_global_desc);
-}
-
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw_no_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_global_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_global_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_global_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_global_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_global_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_global_desc.GetLength(I1);
-
-    const auto Ho = out_n_k_ho_wo_global_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_global_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_global_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_global_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    assert(InLeftPadH == 0 && InLeftPadW == 0 && InRightPadH == 0 && InRightPadW == 0);
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_n_c_y_ho_x_wo_global_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_global_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_global_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_global_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_global_desc, in_gemmk_gemmn_global_desc, out_gemmm_gemmn_global_desc);
-}
-
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw_1x1(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_global_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_global_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_global_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_global_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_global_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_global_desc.GetLength(I1);
-
-    const auto Ho = out_n_k_ho_wo_global_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_global_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_global_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_global_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    assert(Y == 1 && X == 1 && ConvStrideH == 1 && ConvStrideW == 1 && ConvDilationH == 1 &&
-           ConvDilationW == 1 && InLeftPadH == 0 && InLeftPadW == 0 && InRightPadH == 0 &&
-           InRightPadW == 0);
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_gemmk_gemmn_global_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_global_desc,
-        make_tuple(make_pass_through_transform(C), make_merge_transform(make_tuple(N, Ho, Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2, 3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_global_desc, in_gemmk_gemmn_global_desc, out_gemmm_gemmn_global_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index ebfaabb03ebdda41d086e4b80c5856156810992a..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,179 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_grid_desc, in_gemmk_gemmn_grid_desc, out_gemmm_gemmn_grid_desc);
-}
-
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk_1x1(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    assert(Y == 1 && X == 1 && ConvStrideH == 1 && ConvStrideW == 1 && ConvDilationH == 1 &&
-           ConvDilationW == 1 && InLeftPadH == 0 && InLeftPadW == 0 && InRightPadH == 0 &&
-           InRightPadW == 0);
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_gemmk_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, C)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_grid_desc, in_gemmk_gemmn_grid_desc, out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index 6e576d69f5f4b5a9a60a05f84edb43fec990e91a..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,132 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = K;
-    const auto GemmN  = N * Ho * Wo;
-    const auto GemmK  = C * Y * X;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(wei_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(wei_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmk0_gemmn_gemmk1_grid_desc,
-                      out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 13e1bf251abfdc3716bdd17f331f82fd7bdfad41..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,132 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = K;
-    const auto GemmN  = N * Ho * Wo;
-    const auto GemmK  = C * Y * X;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(wei_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(wei_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmk0_gemmn_gemmk1_grid_desc,
-                      out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 088d14b2ee472bc049385ad27e45d02df1e3b6c9..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,134 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: in
-// B: wei
-// C: out
-// GemmM = N * Ho * Wo
-// GemmN = K
-// GemmK = Y * X * C
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = N * Ho * Wo;
-    const auto GemmN  = K;
-    const auto GemmK  = Y * X * C;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // A: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmm_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // B: weight tensor
-    const auto wei_gemmk_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(wei_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // C: output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(in_gemmk0_gemmm_gemmk1_grid_desc,
-                      wei_gemmk0_gemmn_gemmk1_grid_desc,
-                      out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index a6785d56df7e70fdb4882f0c431b886d8c87caac..0000000000000000000000000000000000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,135 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM0 = 1
-// GemmM1 = K
-// GemmN0 = N0
-// GemmN1 = (N / N0) * Ho * Wo
-// GemmK0 = (C / C0) * Y * X
-// GemmK1 = C0
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          typename N0Type,
-          typename C0Type>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_contraction_v6r1_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    const N0Type& N0,
-    const C0Type& C0)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto N1 = N / N0;
-    const auto C1 = C / C0;
-
-    // weight tensor
-    const auto wei_gk0_gm0_gm1_gk1_grid_desc =
-        transform_tensor_descriptor(make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-                                    make_tuple(make_unmerge_transform(make_tuple(I1, K)),
-                                               make_unmerge_transform(make_tuple(C0, C1 * Y * X))),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<1, 2>{}, Sequence<3, 0>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n0_n1_c0_c1_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(N0, N1)),
-                   make_unmerge_transform(make_tuple(C0, C1)),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}, Sequence<6, 7>{}));
-
-    const auto in_gk0_gn0_gn1_gk1_grid_desc = transform_tensor_descriptor(
-        in_n0_n1_c0_c1_y_ho_x_wo_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(C1, Y, X)),
-                   make_pass_through_transform(N0),
-                   make_merge_transform(make_tuple(N1, Ho, Wo)),
-                   make_pass_through_transform(C0)),
-        make_tuple(Sequence<3, 4, 6>{}, Sequence<0>{}, Sequence<1, 5, 7>{}, Sequence<2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    // output tensor
-    const auto out_n_k_howo_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo));
-
-    const auto out_n0_n1_1_k_howo_grid_desc =
-        transform_tensor_descriptor(out_n_k_howo_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(N0, N1)),
-                                               make_unmerge_transform(make_tuple(I1, K)),
-                                               make_pass_through_transform(Ho * Wo)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                                    make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-    const auto out_gm0_gm1_gn0_gn1_grid_desc = transform_tensor_descriptor(
-        out_n0_n1_1_k_howo_grid_desc,
-        make_tuple(make_pass_through_transform(I1),
-                   make_pass_through_transform(K),
-                   make_pass_through_transform(N0),
-                   make_merge_transform_v2_magic_division(make_tuple(N1, Ho * Wo))),
-        make_tuple(Sequence<2>{}, Sequence<3>{}, Sequence<0>{}, Sequence<1, 4>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    return make_tuple(
-        wei_gk0_gm0_gm1_gk1_grid_desc, in_gk0_gn0_gn1_gk1_grid_desc, out_gm0_gm1_gn0_gn1_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/stream_config.hpp b/include/ck/stream_config.hpp
index 70ca34555a01436c79ba244ac03572bb4e9520b4..505a602b240428bcd1f0f81018fef0c4716c80b7 100644
--- a/include/ck/stream_config.hpp
+++ b/include/ck/stream_config.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor/static_tensor.hpp b/include/ck/tensor/static_tensor.hpp
index fee679f91060aca623ab498362de78fd8118fee2..d719ef9760d79297600d7524167eba78cd137831 100644
--- a/include/ck/tensor/static_tensor.hpp
+++ b/include/ck/tensor/static_tensor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_STATIC_TENSOR_HPP
 #define CK_STATIC_TENSOR_HPP
diff --git a/include/ck/tensor_description/cluster_descriptor.hpp b/include/ck/tensor_description/cluster_descriptor.hpp
index 0c9ea2ff2a0d73b793008a954eaf7293b33ade08..2dfcad8e042e548d15b4bd963fe615e39c04eff1 100644
--- a/include/ck/tensor_description/cluster_descriptor.hpp
+++ b/include/ck/tensor_description/cluster_descriptor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/multi_index_transform.hpp b/include/ck/tensor_description/multi_index_transform.hpp
index 4e4d7593e9083f321f1cabe65faaee3c14259666..6854226dd4da975a8df4e5d516232f69476f515f 100644
--- a/include/ck/tensor_description/multi_index_transform.hpp
+++ b/include/ck/tensor_description/multi_index_transform.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/multi_index_transform_helper.hpp b/include/ck/tensor_description/multi_index_transform_helper.hpp
index 044a90370095eb53b94b5c2fba81abdbeae82c00..af0a8a34d0e48bde494dacfff5bbb17eda5eb479 100644
--- a/include/ck/tensor_description/multi_index_transform_helper.hpp
+++ b/include/ck/tensor_description/multi_index_transform_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_adaptor.hpp b/include/ck/tensor_description/tensor_adaptor.hpp
index d42e0a6ff08f60eb1cf6e0f241f93e528bf3514b..3ffac32469a8d97d678bc6b5fe62a7cc5e0b24ab 100644
--- a/include/ck/tensor_description/tensor_adaptor.hpp
+++ b/include/ck/tensor_description/tensor_adaptor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_descriptor.hpp b/include/ck/tensor_description/tensor_descriptor.hpp
index f07d5b1733d9cc96dfac9f18cbbe30510760cabc..f1df2eedd466c81e4b7938c4808075869e6309fd 100644
--- a/include/ck/tensor_description/tensor_descriptor.hpp
+++ b/include/ck/tensor_description/tensor_descriptor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_descriptor_helper.hpp b/include/ck/tensor_description/tensor_descriptor_helper.hpp
index 461aae72cf7b1879c90e473adb3814e1c4875b52..f3ac041bf9b8dc03802d170fbd7bf08ce6ab9cb1 100644
--- a/include/ck/tensor_description/tensor_descriptor_helper.hpp
+++ b/include/ck/tensor_description/tensor_descriptor_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_space_filling_curve.hpp b/include/ck/tensor_description/tensor_space_filling_curve.hpp
index 17c9100b9fd76418d562c5c175ad24062d8c3415..9a326092d2e0fd8392ec42a8c0a82b4167076373 100644
--- a/include/ck/tensor_description/tensor_space_filling_curve.hpp
+++ b/include/ck/tensor_description/tensor_space_filling_curve.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
index 8b1b7be11ef7b3cd19f11bd400b8a96146921bf7..b3caa3214a80893ad1acbf7b5f385bcbc513670b 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp
index 33120bd86ff01da47f7b80c593bc966785cb3711..b0143366c1de5f8e6ec52cceefd3c20e4e77cbb7 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_BLOCKWISE_GEMM_DLOPS_V2R2_HPP
 #define CK_BLOCKWISE_GEMM_DLOPS_V2R2_HPP
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp
index f45655721fe4de11b50b91e0dc5d22790ff73bea..0d092da5168c5d5629eecc2aac7cbdd53e210277 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_BLOCKWISE_GEMM_DLOPS_V3_HPP
 #define CK_BLOCKWISE_GEMM_DLOPS_V3_HPP
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..5ec964bd3fdbda9319d1c6eab0ccb207775a6137
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
@@ -0,0 +1,801 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/warp/wmma_gemm.hpp"
+#include "ck/tensor_description/tensor_adaptor.hpp"
+
+#define CK_MNK_LOOP
+
+namespace ck {
+
+template <index_t BlockSize,
+          typename FloatA,
+          typename FloatB,
+          typename FloatAcc,
+          typename AK0MK1BlockDesc,
+          typename BK0NK1BlockDesc,
+          index_t MPerWMMA,
+          index_t NPerWMMA,
+          index_t MRepeat,
+          index_t NRepeat,
+          index_t KPack>
+/* A: K0PerBlock x MPerBlock x K1
+ * B: K0PerBlock x NPerBlock x K1
+ * C: MRepeat x MWave x MSubGroup x NRepeat x NWave x NThreadPerSubGroup x MAccVgprs
+ * KPACK == WMMA_K = 16
+ */
+struct BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle
+{
+    static constexpr auto I0    = Number<0>{};
+    static constexpr auto I1    = Number<1>{};
+    static constexpr auto I2    = Number<2>{};
+    static constexpr auto I3    = Number<3>{};
+    static constexpr auto I4    = Number<4>{};
+    static constexpr auto WmmaK = Number<16>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    // Hardcode of WaveSize, since current HIP Runtime(5.4.0-10984) could not return correct one.
+    static constexpr index_t WaveSize = 32;
+
+    static constexpr index_t MPerBlock = AK0MK1BlockDesc{}.GetLength(I1);
+    static constexpr index_t NPerBlock = BK0NK1BlockDesc{}.GetLength(I1);
+    static constexpr index_t KPerBlock =
+        BK0NK1BlockDesc{}.GetLength(I0) * BK0NK1BlockDesc{}.GetLength(I2);
+
+    static constexpr index_t A_K0 = AK0MK1BlockDesc{}.GetLength(I0);
+    static constexpr index_t B_K0 = BK0NK1BlockDesc{}.GetLength(I0);
+    static constexpr index_t A_K1 = AK0MK1BlockDesc{}.GetLength(I2);
+    static constexpr index_t B_K1 = BK0NK1BlockDesc{}.GetLength(I2);
+
+    static constexpr auto wmma_gemm =
+        WmmaGemm<FloatA, FloatB, FloatAcc, MPerWMMA, NPerWMMA, KPack>{};
+
+    static constexpr index_t MWaves = MPerBlock / (MRepeat * MPerWMMA);
+    static constexpr index_t NWaves = NPerBlock / (NRepeat * NPerWMMA);
+
+    StaticBufferTupleOfVector<AddressSpaceEnum::Vgpr,
+                              FloatAcc,
+                              MRepeat * NRepeat,
+                              wmma_gemm.GetRegSizePerWmma(),
+                              true>
+        c_thread_buf_;
+
+    __host__ __device__ constexpr auto& GetCThreadBuffer() { return c_thread_buf_; }
+
+    __device__ static auto GetWaveIdx()
+    {
+        const index_t thread_id = ThisThreadBlock::GetThreadId();
+
+        constexpr auto threadid_to_wave_idx_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_merge_transform(make_tuple(MWaves, NWaves, WaveSize))),
+            make_tuple(Sequence<0, 1, 2>{}),
+            make_tuple(Sequence<0>{}));
+
+        return threadid_to_wave_idx_adaptor.CalculateBottomIndex(make_multi_index(thread_id));
+    }
+
+    __device__ static auto CalculateAThreadOriginDataIndex()
+    {
+        const auto wave_idx = GetWaveIdx();
+
+        const auto waveId_m = wave_idx[I0];
+
+        const auto WMMA_a_idx = wmma_gemm.CalculateAThreadOriginDataIndex();
+        //  |KRepeat   |MRepeat|MWave      |MLane       |KPack
+        return make_tuple(0, 0, waveId_m, WMMA_a_idx, 0);
+    }
+
+    __device__ static auto CalculateBThreadOriginDataIndex()
+    {
+        const auto wave_idx = GetWaveIdx();
+
+        const auto waveId_n = wave_idx[I1];
+
+        const auto WMMA_b_idx = wmma_gemm.CalculateBThreadOriginDataIndex();
+        //  |KRepeat   |NRepeat|Nwave      |NLane       |KPack
+        return make_tuple(0, 0, waveId_n, WMMA_b_idx, 0);
+    }
+
+    template <index_t m0, index_t n0>
+    __device__ static auto CalculateCThreadOriginDataIndex(Number<m0>, Number<n0>)
+    {
+        const auto wave_idx = GetWaveIdx();
+
+        const auto waveId_m = wave_idx[I0];
+        const auto waveId_n = wave_idx[I1];
+
+        const auto blk_idx = wmma_gemm.GetBeginOfThreadBlk();
+
+        constexpr auto mrepeat_mwave_mperWMMA_to_m_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_unmerge_transform(make_tuple(MRepeat, MWaves, MPerWMMA))),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0, 1, 2>{}));
+
+        constexpr auto nrepeat_nwave_nperWMMA_to_n_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_unmerge_transform(make_tuple(NRepeat, NWaves, NPerWMMA))),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0, 1, 2>{}));
+
+        const index_t c_thread_m = mrepeat_mwave_mperWMMA_to_m_adaptor.CalculateBottomIndex(
+            make_tuple(m0, waveId_m, blk_idx[I0]))[I0];
+        const index_t c_thread_n = nrepeat_nwave_nperWMMA_to_n_adaptor.CalculateBottomIndex(
+            make_tuple(n0, waveId_n, blk_idx[I1]))[I0];
+
+        return make_tuple(c_thread_m, c_thread_n);
+    }
+
+    __host__ __device__ BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle()
+    {
+        static_assert(AK0MK1BlockDesc::IsKnownAtCompileTime() &&
+                          BK0NK1BlockDesc::IsKnownAtCompileTime(),
+                      "wrong! Desc should be known at compile-time");
+
+        static_assert(ThisThreadBlock::GetNumOfThread() == MWaves * NWaves * WaveSize,
+                      "ThisThreadBlock::GetNumOfThread() != MWaves * NWaves * WaveSize\n");
+
+        static_assert(MPerBlock % (MPerWMMA * MRepeat) == 0 &&
+                          NPerBlock % (NPerWMMA * NRepeat) == 0,
+                      "wrong!");
+    }
+
+    // Thread level, register decriptor. Vector-write
+    __host__ __device__ static constexpr auto
+    GetCThreadDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs()
+    {
+        constexpr auto c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens =
+            wmma_gemm.GetCMSubGroupNThreadPerSubGroupMAccVgprsThreadBlkLengths();
+
+        constexpr auto MSubGroup          = c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens[I0];
+        constexpr auto NThreadPerSubGroup = c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens[I1];
+        constexpr auto MAccVgprs          = c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens[I2];
+
+        return make_naive_tensor_descriptor_packed(
+            //        |MRepeat           |MWave |MSubGroup |NRepeat           |NWave
+            //        |NThreadPerSubGroup |MAccVgprs
+            make_tuple(Number<MRepeat>{},
+                       I1,
+                       MSubGroup,
+                       Number<NRepeat>{},
+                       I1,
+                       NThreadPerSubGroup,
+                       MAccVgprs));
+    }
+
+    // Provide dimension size
+    __host__ __device__ static constexpr auto
+    GetCBlockDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs()
+    {
+        constexpr auto c_block_desc_mrepeat_mwave_mperwmma_nrepeat_nwave_nperwmma =
+            make_naive_tensor_descriptor_packed(make_tuple(Number<MRepeat>{},
+                                                           Number<MWaves>{},
+                                                           Number<MPerWMMA>{},
+                                                           Number<NRepeat>{},
+                                                           Number<NWaves>{},
+                                                           Number<NPerWMMA>{}));
+
+        return wmma_gemm
+            .MakeCDesc_MBlockxRepeat_MWave_MSubGroup_NBlockxRepeat_NWave_NThreadPerSubGroup_MAccVgprs(
+                c_block_desc_mrepeat_mwave_mperwmma_nrepeat_nwave_nperwmma);
+    }
+
+    __host__ __device__ static constexpr auto MakeABlockDescriptor_K0_M0_M1_M2_K1()
+    {
+        return transform_tensor_descriptor(
+            AK0MK1BlockDesc{},
+            make_tuple(make_pass_through_transform(Number<A_K0>{}),
+                       make_unmerge_transform(
+                           make_tuple(Number<MRepeat>{}, Number<MWaves>{}, Number<MPerWMMA>{})),
+                       make_pass_through_transform(Number<A_K1>{})),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+            make_tuple(Sequence<0>{}, Sequence<1, 2, 3>{}, Sequence<4>{}));
+    }
+
+    __host__ __device__ static constexpr auto MakeBBlockDescriptor_K0_N0_N1_N2_K1()
+    {
+        return transform_tensor_descriptor(
+            BK0NK1BlockDesc{},
+            make_tuple(make_pass_through_transform(Number<B_K0>{}),
+                       make_unmerge_transform(
+                           make_tuple(Number<NRepeat>{}, Number<NWaves>{}, Number<NPerWMMA>{})),
+                       make_pass_through_transform(Number<B_K1>{})),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+            make_tuple(Sequence<0>{}, Sequence<1, 2, 3>{}, Sequence<4>{}));
+    }
+
+    // M0_M1_M2 = MRepeat_MWave_MPerWmma, N0_N1_N2 = NRepeat_NWave_NPerWmma
+    static constexpr auto a_block_desc_k0_m0_m1_m2_k1 = MakeABlockDescriptor_K0_M0_M1_M2_K1();
+    static constexpr auto b_block_desc_k0_n0_n1_n2_k1 = MakeBBlockDescriptor_K0_N0_N1_N2_K1();
+
+    template <typename ABlockBuffer, typename BBlockBuffer, typename CThreadBuffer>
+    __device__ void Run(const ABlockBuffer& a_block_buf,
+                        const BBlockBuffer& b_block_buf,
+                        CThreadBuffer& c_thread_buf) const
+    {
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatA>(
+            a_thread_desc_.GetElementSpaceSize());
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatB>(
+            b_thread_desc_.GetElementSpaceSize());
+
+        static_for<0, KPerBlock / WmmaK, 1>{}([&](auto k) { // k=0,1,2 instead of k=0,kpack*1, ...
+            static_for<0, MRepeat, 1>{}([&](auto m0) {
+                // read A
+                a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
+                                   make_tuple(Number<k * WmmaK / A_K1>{}, m0, I0, I0, I0),
+                                   a_block_buf,
+                                   a_thread_desc_,
+                                   make_tuple(I0, m0, I0, I0, I0),
+                                   a_thread_buf);
+
+                static_for<0, NRepeat, 1>{}([&](auto n0) {
+                    // read B
+                    b_thread_copy_.Run(b_block_desc_k0_n0_n1_n2_k1,
+                                       make_tuple(Number<k * WmmaK / B_K1>{}, n0, I0, I0, I0),
+                                       b_block_buf,
+                                       b_thread_desc_,
+                                       make_tuple(I0, n0, I0, I0, I0),
+                                       b_thread_buf);
+                    vector_type<FloatA, WmmaK> a_thread_vec;
+                    vector_type<FloatB, WmmaK> b_thread_vec;
+
+                    static_for<0, WmmaK, 1>{}([&](auto i) {
+                        a_thread_vec.template AsType<FloatA>()(i) =
+                            a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                make_tuple(i / A_K1, m0, 0, 0, i % A_K1))>{}];
+                        b_thread_vec.template AsType<FloatB>()(i) =
+                            b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                make_tuple(i / B_K1, n0, 0, 0, i % B_K1))>{}];
+                    });
+
+                    using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                    using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                    constexpr index_t c_offset =
+                        c_thread_desc_.CalculateOffset(make_tuple(m0, n0, 0));
+
+                    wmma_gemm.template Run(
+                        a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                        b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                        c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                });
+            });
+        });
+    }
+
+    protected:
+    // A[K0, M0, M1, M2, K1]
+    static constexpr auto a_thread_desc_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<WmmaK / A_K1>{}, Number<MRepeat>{}, I1, I1, Number<A_K1>{}));
+
+    // B[K0, N0, N1, N2, K1]
+    static constexpr auto b_thread_desc_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<WmmaK / B_K1>{}, Number<NRepeat>{}, I1, I1, Number<B_K1>{}));
+
+    // C[M, N, NumRegWMMA]
+    static constexpr auto c_thread_desc_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MRepeat>{}, Number<NRepeat>{}, wmma_gemm.GetRegSizePerWmma()));
+
+    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatA,
+                                                         FloatA,
+                                                         decltype(a_block_desc_k0_m0_m1_m2_k1),
+                                                         decltype(a_thread_desc_),
+                                                         Sequence<WmmaK / A_K1, 1, 1, 1, A_K1>,
+                                                         Sequence<0, 1, 2, 3, 4>,
+                                                         4,
+                                                         A_K1,
+                                                         A_K1>;
+
+    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatB,
+                                                         FloatB,
+                                                         decltype(b_block_desc_k0_n0_n1_n2_k1),
+                                                         decltype(b_thread_desc_),
+                                                         Sequence<WmmaK / B_K1, 1, 1, 1, B_K1>,
+                                                         Sequence<0, 1, 2, 3, 4>,
+                                                         4,
+                                                         B_K1,
+                                                         B_K1>;
+
+    AThreadCopy a_thread_copy_{CalculateAThreadOriginDataIndex()};
+    BThreadCopy b_thread_copy_{CalculateBThreadOriginDataIndex()};
+};
+
+// block wise level pipe designed for inline asm
+template <index_t BlockSize,
+          typename FloatA,
+          typename FloatB,
+          typename FloatAcc,
+          typename AK0MK1BlockDesc,
+          typename BK0NK1BlockDesc,
+          index_t MPerWMMA,
+          index_t NPerWMMA,
+          index_t MRepeat,
+          index_t NRepeat,
+          index_t KPack>
+/* A: K0PerBlock x MPerBlock x K1
+ * B: K0PerBlock x NPerBlock x K1
+ * C: MRepeat x MWave x MSubGroup x NRepeat x NWave x NThreadPerSubGroup x MAccVgprs
+ * KPACK == WMMA_K = 16
+ */
+struct BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle_FIFO
+{
+    static constexpr auto I0    = Number<0>{};
+    static constexpr auto I1    = Number<1>{};
+    static constexpr auto I2    = Number<2>{};
+    static constexpr auto I3    = Number<3>{};
+    static constexpr auto I4    = Number<4>{};
+    static constexpr auto WmmaK = Number<16>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    // Hardcode of WaveSize, since current HIP Runtime(5.4.0-10984) could not return correct one.
+    static constexpr index_t WaveSize = 32;
+
+    static constexpr index_t MPerBlock = AK0MK1BlockDesc{}.GetLength(I1);
+    static constexpr index_t NPerBlock = BK0NK1BlockDesc{}.GetLength(I1);
+    static constexpr index_t KPerBlock =
+        BK0NK1BlockDesc{}.GetLength(I0) * BK0NK1BlockDesc{}.GetLength(I2);
+
+    static constexpr index_t A_K0 = AK0MK1BlockDesc{}.GetLength(I0);
+    static constexpr index_t B_K0 = BK0NK1BlockDesc{}.GetLength(I0);
+    static constexpr index_t A_K1 = AK0MK1BlockDesc{}.GetLength(I2);
+    static constexpr index_t B_K1 = BK0NK1BlockDesc{}.GetLength(I2);
+
+    static constexpr auto wmma_gemm =
+        WmmaGemm<FloatA, FloatB, FloatAcc, MPerWMMA, NPerWMMA, KPack>{};
+
+    static constexpr index_t MWaves = MPerBlock / (MRepeat * MPerWMMA);
+    static constexpr index_t NWaves = NPerBlock / (NRepeat * NPerWMMA);
+
+    StaticBufferTupleOfVector<AddressSpaceEnum::Vgpr,
+                              FloatAcc,
+                              MRepeat * NRepeat,
+                              wmma_gemm.GetRegSizePerWmma(),
+                              true>
+        c_thread_buf_;
+
+    __host__ __device__ constexpr auto& GetCThreadBuffer() { return c_thread_buf_; }
+
+    __device__ static auto GetWaveIdx()
+    {
+        const index_t thread_id = ThisThreadBlock::GetThreadId();
+
+        constexpr auto threadid_to_wave_idx_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_merge_transform(make_tuple(MWaves, NWaves, WaveSize))),
+            make_tuple(Sequence<0, 1, 2>{}),
+            make_tuple(Sequence<0>{}));
+
+        return threadid_to_wave_idx_adaptor.CalculateBottomIndex(make_multi_index(thread_id));
+    }
+
+    __device__ static auto CalculateAThreadOriginDataIndex()
+    {
+        const auto wave_idx = GetWaveIdx();
+
+        const auto waveId_m = wave_idx[I0];
+
+        const auto WMMA_a_idx = wmma_gemm.CalculateAThreadOriginDataIndex();
+        //  |KRepeat   |MRepeat|MWave      |MLane       |KPack
+        return make_tuple(0, 0, waveId_m, WMMA_a_idx, 0);
+    }
+
+    __device__ static auto CalculateBThreadOriginDataIndex()
+    {
+        const auto wave_idx = GetWaveIdx();
+
+        const auto waveId_n = wave_idx[I1];
+
+        const auto WMMA_b_idx = wmma_gemm.CalculateBThreadOriginDataIndex();
+        //  |KRepeat   |NRepeat|Nwave      |NLane       |KPack
+        return make_tuple(0, 0, waveId_n, WMMA_b_idx, 0);
+    }
+
+    template <index_t m0, index_t n0>
+    __device__ static auto CalculateCThreadOriginDataIndex(Number<m0>, Number<n0>)
+    {
+        const auto wave_idx = GetWaveIdx();
+
+        const auto waveId_m = wave_idx[I0];
+        const auto waveId_n = wave_idx[I1];
+
+        const auto blk_idx = wmma_gemm.GetBeginOfThreadBlk();
+
+        constexpr auto mrepeat_mwave_mperWMMA_to_m_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_unmerge_transform(make_tuple(MRepeat, MWaves, MPerWMMA))),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0, 1, 2>{}));
+
+        constexpr auto nrepeat_nwave_nperWMMA_to_n_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_unmerge_transform(make_tuple(NRepeat, NWaves, NPerWMMA))),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0, 1, 2>{}));
+
+        const index_t c_thread_m = mrepeat_mwave_mperWMMA_to_m_adaptor.CalculateBottomIndex(
+            make_tuple(m0, waveId_m, blk_idx[I0]))[I0];
+        const index_t c_thread_n = nrepeat_nwave_nperWMMA_to_n_adaptor.CalculateBottomIndex(
+            make_tuple(n0, waveId_n, blk_idx[I1]))[I0];
+
+        return make_tuple(c_thread_m, c_thread_n);
+    }
+
+    __host__ __device__ BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle_FIFO()
+    {
+        static_assert(AK0MK1BlockDesc::IsKnownAtCompileTime() &&
+                          BK0NK1BlockDesc::IsKnownAtCompileTime(),
+                      "wrong! Desc should be known at compile-time");
+
+        static_assert(ThisThreadBlock::GetNumOfThread() == MWaves * NWaves * WaveSize,
+                      "ThisThreadBlock::GetNumOfThread() != MWaves * NWaves * WaveSize\n");
+
+        static_assert(MPerBlock % (MPerWMMA * MRepeat) == 0 &&
+                          NPerBlock % (NPerWMMA * NRepeat) == 0,
+                      "wrong!");
+    }
+    // Thread level, register decriptor. Vector-write
+    __host__ __device__ static constexpr auto
+    GetCThreadDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs()
+    {
+        constexpr auto c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens =
+            wmma_gemm.GetCMSubGroupNThreadPerSubGroupMAccVgprsThreadBlkLengths();
+
+        constexpr auto MSubGroup          = c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens[I0];
+        constexpr auto NThreadPerSubGroup = c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens[I1];
+        constexpr auto MAccVgprs          = c_msubgroup_nthreadpersubgroup_maccvgprs_tblk_lens[I2];
+
+        return make_naive_tensor_descriptor_packed(
+            //        |MRepeat           |MWave |MSubGroup |NRepeat           |NWave
+            //        |NThreadPerSubGroup |MAccVgprs
+            make_tuple(Number<MRepeat>{},
+                       I1,
+                       MSubGroup,
+                       Number<NRepeat>{},
+                       I1,
+                       NThreadPerSubGroup,
+                       MAccVgprs));
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeCGridDescriptor_MBlockxRepeat_MWave_MSubGroup_NBlockxRepeat_NWave_NThreadPerSubGroup_MAccVgprs(
+        const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        const auto M = c_grid_desc_m_n.GetLength(I0);
+        const auto N = c_grid_desc_m_n.GetLength(I1);
+
+        const auto c_grid_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma =
+            transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(M / (MWaves * MPerWMMA), MWaves, MPerWMMA)),
+                    make_unmerge_transform(make_tuple(N / (NWaves * NPerWMMA), NWaves, NPerWMMA))),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 2>{}, Sequence<3, 4, 5>{}));
+
+        return wmma_gemm
+            .MakeCDesc_MBlockxRepeat_MWave_MSubGroup_NBlockxRepeat_NWave_NThreadPerSubGroup_MAccVgprs(
+                c_grid_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma);
+    }
+
+    // Provide dimension size
+    __host__ __device__ static constexpr auto
+    GetCBlockDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs()
+    {
+        constexpr auto c_block_desc_mrepeat_mwave_mperwmma_nrepeat_nwave_nperwmma =
+            make_naive_tensor_descriptor_packed(make_tuple(Number<MRepeat>{},
+                                                           Number<MWaves>{},
+                                                           Number<MPerWMMA>{},
+                                                           Number<NRepeat>{},
+                                                           Number<NWaves>{},
+                                                           Number<NPerWMMA>{}));
+
+        return wmma_gemm
+            .MakeCDesc_MBlockxRepeat_MWave_MSubGroup_NBlockxRepeat_NWave_NThreadPerSubGroup_MAccVgprs(
+                c_block_desc_mrepeat_mwave_mperwmma_nrepeat_nwave_nperwmma);
+    }
+
+    __host__ __device__ static constexpr auto MakeABlockDescriptor_K0_M0_M1_M2_K1()
+    {
+        return transform_tensor_descriptor(
+            AK0MK1BlockDesc{},
+            make_tuple(make_pass_through_transform(Number<A_K0>{}),
+                       make_unmerge_transform(
+                           make_tuple(Number<MRepeat>{}, Number<MWaves>{}, Number<MPerWMMA>{})),
+                       make_pass_through_transform(Number<A_K1>{})),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+            make_tuple(Sequence<0>{}, Sequence<1, 2, 3>{}, Sequence<4>{}));
+    }
+
+    __host__ __device__ static constexpr auto MakeBBlockDescriptor_K0_N0_N1_N2_K1()
+    {
+        return transform_tensor_descriptor(
+            BK0NK1BlockDesc{},
+            make_tuple(make_pass_through_transform(Number<B_K0>{}),
+                       make_unmerge_transform(
+                           make_tuple(Number<NRepeat>{}, Number<NWaves>{}, Number<NPerWMMA>{})),
+                       make_pass_through_transform(Number<B_K1>{})),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+            make_tuple(Sequence<0>{}, Sequence<1, 2, 3>{}, Sequence<4>{}));
+    }
+
+    // M0_M1_M2 = MRepeat_MWave_MPerWmma, N0_N1_N2 = NRepeat_NWave_NPerWmma
+    static constexpr auto a_block_desc_k0_m0_m1_m2_k1 = MakeABlockDescriptor_K0_M0_M1_M2_K1();
+    static constexpr auto b_block_desc_k0_n0_n1_n2_k1 = MakeBBlockDescriptor_K0_N0_N1_N2_K1();
+
+    template <typename ABlockBuffer, typename BBlockBuffer, typename CThreadBuffer>
+    __device__ void Run(const ABlockBuffer& a_block_buf,
+                        const BBlockBuffer& b_block_buf,
+                        CThreadBuffer& c_thread_buf) const
+    {
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatA>(
+            a_thread_desc_.GetElementSpaceSize());
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatB>(
+            b_thread_desc_.GetElementSpaceSize());
+
+        constexpr auto RepeatDiff = MRepeat - NRepeat;
+        // Read all Mrepeat, Nrepeat
+        static_for<0, NRepeat, 1>{}([&](auto iN) {
+            b_thread_copy_.Run(b_block_desc_k0_n0_n1_n2_k1,
+                               make_tuple(I0, Number<iN>{}, I0, I0, I0),
+                               b_block_buf,
+                               b_thread_desc_,
+                               make_tuple(I0, Number<iN>{}, I0, I0, I0),
+                               b_thread_buf);
+        });
+
+        static_for<0, MRepeat, 1>{}([&](auto iM) {
+            a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
+                               make_tuple(I0, Number<iM>{}, I0, I0, I0),
+                               a_block_buf,
+                               a_thread_desc_,
+                               make_tuple(I0, Number<iM>{}, I0, I0, I0),
+                               a_thread_buf);
+        });
+
+        // Stage 1: Cut to Repeat Retangle to Square, assume MRepeat > NRepeat
+        static_for<0, RepeatDiff, 1>{}([&](auto iCut) {
+            static_for<0, NRepeat, 1>{}([&](auto iN) {
+                vector_type<FloatA, WmmaK> a_thread_vec;
+                vector_type<FloatB, WmmaK> b_thread_vec;
+
+                static_for<0, WmmaK, 1>{}([&](auto iK) {
+                    a_thread_vec.template AsType<FloatA>()(iK) =
+                        a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                            make_tuple(iK / A_K1, iCut, 0, 0, iK % A_K1))>{}];
+                    b_thread_vec.template AsType<FloatB>()(iK) =
+                        b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                            make_tuple(iK / B_K1, iN, 0, 0, iK % B_K1))>{}];
+                });
+                using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                constexpr index_t c_offset =
+                    c_thread_desc_.CalculateOffset(make_tuple(iCut, iN, 0));
+                // s_nop();
+                wmma_gemm.template Run(
+                    a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                    b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                    c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                // s_nop();
+            });
+            if constexpr(KPerBlock > WmmaK)
+            {
+                // Read Consumed Next inner loop A
+                a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
+                                   make_tuple(Number<WmmaK / A_K1>{}, Number<iCut>{}, I0, I0, I0),
+                                   a_block_buf,
+                                   a_thread_desc_,
+                                   make_tuple(I0, Number<iCut>{}, I0, I0, I0),
+                                   a_thread_buf);
+            }
+        });
+
+        static_for<WmmaK, KPerBlock, WmmaK>{}([&](auto iWmmaK) {
+            // Stage 2: Run FIFO fashion loopover in Square
+            static_for<0, NRepeat, 1>{}([&](auto WmmaInnerloop) {
+                // Row Repeatation
+                static_for<WmmaInnerloop, NRepeat, 1>{}([&](auto iN) {
+                    vector_type<FloatA, WmmaK> a_thread_vec;
+                    vector_type<FloatB, WmmaK> b_thread_vec;
+
+                    static_for<0, WmmaK, 1>{}([&](auto iK) {
+                        a_thread_vec.template AsType<FloatA>()(iK) =
+                            a_thread_buf[Number<a_thread_desc_.CalculateOffset(make_tuple(
+                                iK / A_K1, WmmaInnerloop + RepeatDiff, 0, 0, iK % A_K1))>{}];
+                        b_thread_vec.template AsType<FloatB>()(iK) =
+                            b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                make_tuple(iK / B_K1, iN, 0, 0, iK % B_K1))>{}];
+                    });
+                    using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                    using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                    constexpr index_t c_offset = c_thread_desc_.CalculateOffset(
+                        make_tuple(WmmaInnerloop + RepeatDiff, iN, 0));
+                    // s_nop();
+                    wmma_gemm.template Run(
+                        a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                        b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                        c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                    // s_nop();
+                });
+
+                // Read Consumed Next inner loop A
+                a_thread_copy_.Run(
+                    a_block_desc_k0_m0_m1_m2_k1,
+                    make_tuple(
+                        Number<iWmmaK / A_K1>{}, Number<WmmaInnerloop + RepeatDiff>{}, I0, I0, I0),
+                    a_block_buf,
+                    a_thread_desc_,
+                    make_tuple(I0, Number<WmmaInnerloop + RepeatDiff>{}, I0, I0, I0),
+                    a_thread_buf);
+
+                // Col Repeatation
+                static_for<WmmaInnerloop + 1 + RepeatDiff, MRepeat, 1>{}([&](auto iM) {
+                    vector_type<FloatA, WmmaK> a_thread_vec;
+                    vector_type<FloatB, WmmaK> b_thread_vec;
+
+                    static_for<0, WmmaK, 1>{}([&](auto iK) {
+                        a_thread_vec.template AsType<FloatA>()(iK) =
+                            a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                make_tuple(iK / A_K1, iM, 0, 0, iK % A_K1))>{}];
+                        b_thread_vec.template AsType<FloatB>()(iK) =
+                            b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                make_tuple(iK / B_K1, WmmaInnerloop, 0, 0, iK % B_K1))>{}];
+                    });
+                    using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                    using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                    constexpr index_t c_offset =
+                        c_thread_desc_.CalculateOffset(make_tuple(iM, WmmaInnerloop, 0));
+                    // s_nop();
+                    wmma_gemm.template Run(
+                        a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                        b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                        c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                    // s_nop();
+                });
+                // Read Consumed Next inner loop B
+                b_thread_copy_.Run(
+                    b_block_desc_k0_n0_n1_n2_k1,
+                    make_tuple(Number<iWmmaK / B_K1>{}, Number<WmmaInnerloop>{}, I0, I0, I0),
+                    b_block_buf,
+                    b_thread_desc_,
+                    make_tuple(I0, Number<WmmaInnerloop>{}, I0, I0, I0),
+                    b_thread_buf);
+            });
+
+            // Stage 1: Cut to Repeat Retangle to Square, assume MRepeat > NRepeat
+            static_for<0, RepeatDiff, 1>{}([&](auto iCut) {
+                static_for<0, NRepeat, 1>{}([&](auto iN) {
+                    vector_type<FloatA, WmmaK> a_thread_vec;
+                    vector_type<FloatB, WmmaK> b_thread_vec;
+
+                    static_for<0, WmmaK, 1>{}([&](auto iK) {
+                        a_thread_vec.template AsType<FloatA>()(iK) =
+                            a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                make_tuple(iK / A_K1, iCut, 0, 0, iK % A_K1))>{}];
+                        b_thread_vec.template AsType<FloatB>()(iK) =
+                            b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                make_tuple(iK / B_K1, iN, 0, 0, iK % B_K1))>{}];
+                    });
+                    using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                    using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                    constexpr index_t c_offset =
+                        c_thread_desc_.CalculateOffset(make_tuple(iCut, iN, 0));
+                    // s_nop();
+                    wmma_gemm.template Run(
+                        a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                        b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                        c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                    // s_nop();
+                });
+                if constexpr(KPerBlock > WmmaK)
+                {
+                    a_thread_copy_.Run(
+                        a_block_desc_k0_m0_m1_m2_k1,
+                        make_tuple(Number<(iWmmaK + WmmaK) / A_K1>{}, Number<iCut>{}, I0, I0, I0),
+                        a_block_buf,
+                        a_thread_desc_,
+                        make_tuple(I0, Number<iCut>{}, I0, I0, I0),
+                        a_thread_buf);
+                }
+            });
+        });
+
+        // Stage 2: Run FIFO fashion loopover in Square
+        static_for<0, NRepeat, 1>{}([&](auto WmmaInnerloop) {
+            // Row Repeatation
+            static_for<WmmaInnerloop, NRepeat, 1>{}([&](auto iN) {
+                vector_type<FloatA, WmmaK> a_thread_vec;
+                vector_type<FloatB, WmmaK> b_thread_vec;
+
+                static_for<0, WmmaK, 1>{}([&](auto iK) {
+                    a_thread_vec.template AsType<FloatA>()(iK) =
+                        a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                            make_tuple(iK / A_K1, WmmaInnerloop + RepeatDiff, 0, 0, iK % A_K1))>{}];
+                    b_thread_vec.template AsType<FloatB>()(iK) =
+                        b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                            make_tuple(iK / B_K1, iN, 0, 0, iK % B_K1))>{}];
+                });
+                using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                constexpr index_t c_offset =
+                    c_thread_desc_.CalculateOffset(make_tuple(WmmaInnerloop + RepeatDiff, iN, 0));
+                // s_nop();
+                wmma_gemm.template Run(
+                    a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                    b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                    c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                // s_nop();
+            });
+
+            // Col Repeatation
+            static_for<WmmaInnerloop + 1 + RepeatDiff, MRepeat, 1>{}([&](auto iM) {
+                vector_type<FloatA, WmmaK> a_thread_vec;
+                vector_type<FloatB, WmmaK> b_thread_vec;
+
+                static_for<0, WmmaK, 1>{}([&](auto iK) {
+                    a_thread_vec.template AsType<FloatA>()(iK) =
+                        a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                            make_tuple(iK / A_K1, iM, 0, 0, iK % A_K1))>{}];
+                    b_thread_vec.template AsType<FloatB>()(iK) =
+                        b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                            make_tuple(iK / B_K1, WmmaInnerloop, 0, 0, iK % B_K1))>{}];
+                });
+                using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                constexpr index_t c_offset =
+                    c_thread_desc_.CalculateOffset(make_tuple(iM, WmmaInnerloop, 0));
+                // s_nop();
+                wmma_gemm.template Run(
+                    a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                    b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                    c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                // s_nop();
+            });
+        });
+    }
+
+    protected:
+    // A[M0, M1, M2, K0 = WmmaK]
+    static constexpr auto a_thread_desc_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<WmmaK / A_K1>{}, Number<MRepeat>{}, I1, I1, Number<A_K1>{}));
+
+    // B[N0, N1, N2, K0 = WmmaK]
+    static constexpr auto b_thread_desc_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<WmmaK / B_K1>{}, Number<NRepeat>{}, I1, I1, Number<B_K1>{}));
+
+    // C[M, N, NumRegWMMA]
+    static constexpr auto c_thread_desc_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MRepeat>{}, Number<NRepeat>{}, wmma_gemm.GetRegSizePerWmma()));
+
+    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatA,
+                                                         FloatA,
+                                                         decltype(a_block_desc_k0_m0_m1_m2_k1),
+                                                         decltype(a_thread_desc_),
+                                                         Sequence<WmmaK / A_K1, 1, 1, 1, A_K1>,
+                                                         Sequence<0, 1, 2, 3, 4>,
+                                                         4,
+                                                         A_K1,
+                                                         A_K1>;
+
+    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatB,
+                                                         FloatB,
+                                                         decltype(b_block_desc_k0_n0_n1_n2_k1),
+                                                         decltype(b_thread_desc_),
+                                                         Sequence<WmmaK / B_K1, 1, 1, 1, B_K1>,
+                                                         Sequence<0, 1, 2, 3, 4>,
+                                                         4,
+                                                         B_K1,
+                                                         B_K1>;
+
+    AThreadCopy a_thread_copy_{CalculateAThreadOriginDataIndex()};
+    BThreadCopy b_thread_copy_{CalculateBThreadOriginDataIndex()};
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
index deb26779b22a90e13d362e42e9a35ad7405ed6c8..7d070b03a2cd4ce908a23383b03d07e71e1a02d4 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -705,11 +705,16 @@ constexpr auto BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector()
     }
 };
 
-// Blockwise gemm supporting
-// 1. regular XDL output M2_M3_M4_M2 and transposed XDL output M2_N2_N3_N4
-// 2. decoupled input tile descriptor and mma tile descriptor in order to support both vgpr and LDS
-// source buffer
-// 3. configurable k index starting position and step size after each FMA/XDL instruction
+/**
+ * @brief Blockwise gemm
+ *
+ * Supports
+ * 1. regular XDL output M2_M3_M4_M2 and transposed XDL output M2_N2_N3_N4
+ * 2. decoupled input tile descriptor and mma tile descriptor in order to support both vgpr and LDS
+ * source buffer
+ * 3. configurable k index starting position and step size after each FMA/XDL instruction
+ */
+
 template <index_t BlockSize,
           typename FloatAB,
           typename FloatAcc,
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
index aa814ab00939d0c36866af01b6cf2057dcbd5121..8ae1ba3f34c10d8359b470ed1172f1b70a7fa8b5 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp b/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
index 0dcc82a00a7092b8e51c6653e47bf00d0a6ea47d..8c9ed153eee5fb421dabb015be85c833baf65166 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -12,6 +12,16 @@
 
 namespace ck {
 
+/**
+ * @brief Blockwise softmax
+ *
+ * @tparam BlockSize Block size
+ * @tparam AccDataType Accumulator data type
+ * @tparam ThreadMap_M_K Thread id to m_k
+ * @tparam ThreadClusterDesc_M_K Threadwise cluster descriptor
+ * @tparam ThreadSliceDesc_M_K Threadwise slices descriptor
+ * @tparam IgnoreNaN Flag to ignore NaN, false by default
+ */
 template <index_t BlockSize,
           typename AccDataType,
           typename ThreadMap_M_K, // thread_id to m_k
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp b/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp
index 03e4d42d3a1f9eb1b180c95368b905e619e67110..d8da134a3415a6976b27d0a4fdd7f13798d0245b 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp b/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
index 316508651e4bd4dad2e47edf020a4145176fa753..820a08fc482b44a3509188db0f98aecabb692a82 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck/tensor_description/cluster_descriptor.hpp"
-#include "ck/utility/reduction_common.hpp"
+#include "ck/utility/get_shift.hpp"
 
 namespace ck {
 
@@ -35,10 +35,11 @@ struct BlockwiseWelford
     static constexpr auto thread_cluster_desc =
         make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
 
+    template <typename CountDataType>
     __device__ static inline void
-    Merge(T& mean_a, T& var_a, int& count_a, T mean_b, T var_b, int count_b)
+    Merge(T& mean_a, T& var_a, CountDataType& count_a, T mean_b, T var_b, CountDataType count_b)
     {
-        int count            = count_a + count_b;
+        CountDataType count  = count_a + count_b;
         T count_b_over_count = count == 0 ? type_convert<T>(0) : type_convert<T>(count_b) / count;
         T delta              = mean_b - mean_a;
         mean_a += delta * count_b_over_count;
@@ -46,11 +47,12 @@ struct BlockwiseWelford
         count_a = count;
     }
 
-    __device__ static void Run(T& mean_value, T& var_value, int& count)
+    template <typename CountDataType>
+    __device__ static void Run(T& mean_value, T& var_value, CountDataType& count)
     {
         __shared__ T mean_block_buf[BlockSize];
         __shared__ T var_block_buf[BlockSize];
-        __shared__ int count_block_buf[BlockSize];
+        __shared__ CountDataType count_block_buf[BlockSize];
 
         constexpr auto cluster_len_shift = get_shift<BufferLength_K>();
 
@@ -76,13 +78,13 @@ struct BlockwiseWelford
                 index_t offset2 = block_buf_desc_m_k.CalculateOffset(thread_cluster_idx +
                                                                      make_tuple(0, indOffset));
 
-                T mean1    = mean_block_buf[offset1];
-                T var1     = var_block_buf[offset1];
-                int count1 = count_block_buf[offset1];
+                T mean1              = mean_block_buf[offset1];
+                T var1               = var_block_buf[offset1];
+                CountDataType count1 = count_block_buf[offset1];
 
-                T mean2    = mean_block_buf[offset2];
-                T var2     = var_block_buf[offset2];
-                int count2 = count_block_buf[offset2];
+                T mean2              = mean_block_buf[offset2];
+                T var2               = var_block_buf[offset2];
+                CountDataType count2 = count_block_buf[offset2];
 
                 Merge(mean1, var1, count1, mean2, var2, count2);
 
diff --git a/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp b/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
index 2163ad32383d4f940ab4cf3b041ab2d5c61456cc..82667e235238170806846940e5d153ea20992b73 100644
--- a/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
+++ b/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck/tensor_description/cluster_descriptor.hpp"
-#include "ck/utility/reduction_common.hpp"
+#include "ck/utility/get_shift.hpp"
 #include "ck/utility/reduction_functions_accumulate.hpp"
 
 namespace ck {
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
index 0e5dfb355fb00565f8d9834a2e0df05bfd9db416..c8690e5f68765622222ac3a8cff7c59e6fc06e91 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -11,10 +11,15 @@
 
 namespace ck {
 
-// this version does following things to avoid scratch memory issue
-// 1. Use StaticallyIndexedArray instead of C array for thread buffer
-// 2. ThreadwiseTensorSliceTransfer_v3 does not keep reference to tensor descriptor
-// 3. ThreadwiseTensorSliceTransfer_v3::Run() does not construct new tensor coordinate
+/**
+ * @brief Blockwise data transfer
+ *
+ * This version does following things to avoid scratch memory issue
+ * 1. Use StaticallyIndexedArray instead of C array for thread buffer
+ * 2. ThreadwiseTensorSliceTransfer_v3 does not keep reference to tensor descriptor
+ * 3. ThreadwiseTensorSliceTransfer_v3::Run() does not construct new tensor coordinate
+ *
+ */
 template <typename ThreadGroup,
           typename SrcElementwiseOperation,
           typename DstElementwiseOperation,
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp
index 5c47a49b38b6a30eebc9190cc338d4dbc0bc8524..905a59f56e3b42bfb4b2bdf8652a3e81b2872076 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp
index aa33fc083f15cfd8904c8fa086a866dbc7817e7c..17110c8358dbaf7e602e7f9822968e08f80e14ef 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp
index eb5f589a4ada976c9be4a5001fb6fc288c7e9a43..9a5317dd126a818bc88043408427f0f105cb7f6e 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp
index 3bd7806389b59542b191e61f5369aebf7300fc6a..993d90e356def70ba54e357889277188b4f21971 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp b/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
index a4a29f5d5edc3b840fc7489d5f706231990cad70..f3263c7216a61528bed5408d0a950f8ae40983e1 100644
--- a/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp b/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp
index 20b2a152b9d276b0bd0f6e1aa5798b4817b013a9..01bb806789c3d9e50f706d7c88982b4a6624ff57 100644
--- a/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp b/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp
index 953ff1e06ed07b24968c7f5c0842161ac66643ed..adfa1689c66509c8c194985365356740d4c90473 100644
--- a/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_base.hpp b/include/ck/tensor_operation/gpu/device/device_base.hpp
index 5946daf21ec169612c3207282007e05671882f09..198169011107fb0f236d4657c399ff0534ce2c98 100644
--- a/include/ck/tensor_operation/gpu/device/device_base.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_base.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp
index 9fcd893c7a8c38d31f480b13eb8e7bfc97ec7fef..ee7af0117d17acc561e5943ac59611e188b6ba4c 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp
index e755913280f73e325daf1352324bab8d3df8a3a9..6cc2c7bb2f6c176f2d84fdda4be2140db5564360 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp
index af681127f30404f182475039add4443e46213398..91b4b6b91b6b6be5d08d88e5ca68c9b62fb02aea 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp
index 116e62c00907e0f72031552f46b8a3ea975a18f5..f18dc3290600e63edb869d93dcb1a903206e32ab 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp
index eacc5976d3ea3885a8294f1ee74a249f0a3d52a7..8234e29486b1af8ae0dd43e58d4be9b63b5ed34f 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp
index c1f85e575ce4b2e0d70c05010bfca6e9ed9b84a8..09259224e75cc67e092eeeaab1af793b050cc1f0 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp
index 2b7ab0c9c4f5500d6f90bc871db4e7cb207593c8..5504245c10479c10541fe9053820b2c671fb1a84 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -27,9 +27,9 @@ template <index_t NumDimG,
           typename Acc1BiasDataType,
           typename AElementwiseOperation,
           typename B0ElementwiseOperation,
-          typename Acc0ElementwiseOperation,
+          typename C0DEElementwiseOperation,
           typename B1ElementwiseOperation,
-          typename CElementwiseOperation,
+          typename C1DEElementwiseOperation,
           MaskingSpecialization MaskingSpec>
 struct DeviceBatchedGemmSoftmaxGemmPermute : public BaseOperator
 {
@@ -59,9 +59,9 @@ struct DeviceBatchedGemmSoftmaxGemmPermute : public BaseOperator
             acc1_biases_gs_ms_gemm1ns_strides, // acc1_biases_gs_ms_os_strides
         AElementwiseOperation a_element_op,
         B0ElementwiseOperation b0_element_op,
-        Acc0ElementwiseOperation acc0_element_op,
+        C0DEElementwiseOperation c0de_element_op,
         B1ElementwiseOperation b1_element_op,
-        CElementwiseOperation c_element_op) = 0;
+        C1DEElementwiseOperation c1de_element_op) = 0;
 
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
 };
diff --git a/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp b/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp
index d39f3b7cbcfc9ccaccca5f755be747cc4896dc1b..2c0da692570b455da42540bb72793b9e629cb823 100644
--- a/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp b/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp
index aa93dd9c19d77fbee7ccdafd1e41e4c4f64735cd..e3962e177ee824981c35fbe2ba04664e4124315b 100644
--- a/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp b/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp
index 8a00fd9db3327a7c536b1ada0ca090bfa847a256..69103b6f44297ed65ed4233e3d458ac6927bcd16 100644
--- a/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_cgemm.hpp b/include/ck/tensor_operation/gpu/device/device_cgemm.hpp
index aedae53800b03cb78e6af5b15c8c6dfcd9792eea..8484212118c796f68d1bf066eab7b8ddbb231797 100644
--- a/include/ck/tensor_operation/gpu/device/device_cgemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_cgemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "device_base.hpp"
diff --git a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
index dbc525c099bca293d106ac6d3c37ac72b2d749cc..118ade8978643f10dcfafbefb552fe485d6697e4 100644
--- a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp b/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp
index 82054a3c9423f819d17a003b8c22844f0a2273fe..eb1b85ec822aa96822b5a288d55a5d3a0898f1e9 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp
index 4b9881088dde07a62e92c5f77a19edc9c4b7670f..4dc11dbefd73619c39febe31a91a953b25049f05 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp
index 5a627deeb2221f5e271532d45bc8a544c8cceeba..7d3845666cf2289f34b186a5aaa8ee616812f30a 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp
index cc139303c929ae999d182ad09e9a4ef33f5209e1..3a49ac632e7de745e78ad5670dc57f425ab9eb33 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_elementwise_base.hpp b/include/ck/tensor_operation/gpu/device/device_elementwise.hpp
similarity index 83%
rename from include/ck/tensor_operation/gpu/device/device_elementwise_base.hpp
rename to include/ck/tensor_operation/gpu/device/device_elementwise.hpp
index 728faf543df6cdd5a6593224ee0bb1801b22cd06..db0e4bd83f46ad001492876ff8a712895a873c22 100644
--- a/include/ck/tensor_operation/gpu/device/device_elementwise_base.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_elementwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,7 +17,7 @@ template <typename InDataTypeTuple,
           typename OutDataTypeTuple,
           typename ElementwiseOperation,
           index_t NumDim>
-struct DeviceElementwiseBase : public BaseOperator
+struct DeviceElementwise : public BaseOperator
 {
     static constexpr int NumInput  = InDataTypeTuple::Size();
     static constexpr int NumOutput = OutDataTypeTuple::Size();
@@ -37,8 +37,8 @@ template <typename InDataTypeTuple,
           typename OutDataTypeTuple,
           typename ElementwiseOperation,
           index_t NumDim>
-using DeviceElementwiseBasePtr = std::unique_ptr<
-    DeviceElementwiseBase<InDataTypeTuple, OutDataTypeTuple, ElementwiseOperation, NumDim>>;
+using DeviceElementwisePtr = std::unique_ptr<
+    DeviceElementwise<InDataTypeTuple, OutDataTypeTuple, ElementwiseOperation, NumDim>>;
 
 } // namespace device
 } // namespace tensor_operation
diff --git a/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp b/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp
index d8a791c322b50d28613d1b48f7029539abb2ebf6..c56a947ec9e11a0a5dce2cd39ce86d8e6a3efb8c 100644
--- a/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -32,7 +32,7 @@ struct DeviceElementwiseNormalization : public BaseOperator
                         const std::vector<index_t> betaStrides,
                         const std::vector<index_t> yStrides,
                         const std::vector<index_t> reduceDims,
-                        AccDataType epsilon,
+                        double epsilon,
                         const std::array<const void*, NumInput> in_dev_buffers,
                         const void* p_gamma,
                         const void* p_beta,
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_gemm.hpp
index c0af6f80faf606a22678e516be994a75c1d56eca..adf909821dbbdde12526513a7b6218e8ebbd219d 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp
index 4c2161eaed5b9c7d7913686f44c360c044ca3b07..a7f42c3b35e237cb4ad9edcbcd137f453af59051 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp
index 9113bb7b7454443cce4871bf263433d4f907bfba..a44356dc2405bfe6ed448a8193d55e86bd29415c 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0258858fe50e2d433e5ddfe007e8d06227b47c0c
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp
@@ -0,0 +1,67 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <array>
+#include "device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   output : H[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+//   H = layernorm(E)
+// Assume:
+//   D0, D1, ... and E have the same layout
+//   Calculate mean & variance along N dimension in layernorm(E)
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename HLayout,
+          typename ADataType,
+          typename BDataType,
+          typename DsDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename HDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename HElementwiseOperation>
+struct DeviceGemmMultipleDLayernorm : public BaseOperator
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        const void* p_gamma,
+                        const void* p_beta,
+                        void* p_h,
+                        index_t MRaw,
+                        index_t NRaw,
+                        index_t KRaw,
+                        index_t StrideA,
+                        index_t StrideB,
+                        std::array<index_t, NumDTensor> StrideDs,
+                        index_t StrideH,
+                        double epsilon,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op,
+                        HElementwiseOperation h_element_op) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+}; // namespace device
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp
index f4881e32f620be9c31649085813a257e9e84a598..539e83f7cb8a5dfe99b2ecb3da491f9661330265 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp
index fcc088ca43d1b8c6224c0015e3eb7434038af4b2..eaa7671c6424b4ecf5ad87c4e790a9c478645b9b 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
index c701bff57f8eb7db155e9abdfd3ab7210e7eeffd..6407aa7e09b81a4dc09dcb1bdca5701aa86d9e14 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp
index 173c613a325d8c594298ff6751060fecf2f8d453..ba81948440acd892a9877a7615f95642628b6347 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
index 3350aec8d3174010cf0d926c3a91802ce3b8e7e7..7e4bca2bd66a55c70b6a9365e2f1debcc22bb119 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
index 1258aed71c502c659f065d08e8b1a84922c26c83..a103bdff351ef6983c0435b5b6b06570b00b5eec 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -27,17 +27,19 @@ struct DeviceGroupedConvBwdWeight : public BaseOperator
     MakeArgumentPointer(const void* p_in,
                         void* p_wei,
                         const void* p_out,
-                        ck::index_t G,
-                        ck::index_t N,
-                        ck::index_t K,
-                        ck::index_t C,
-                        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                        std::array<ck::index_t, NDimSpatial> input_left_pads,
-                        std::array<ck::index_t, NDimSpatial> input_right_pads,
+                        const ck::index_t G,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t C,
+                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                         InElementwiseOperation in_element_op,
                         WeiElementwiseOperation wei_element_op,
                         OutElementwiseOperation out_element_op,
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
index 644c7ee9a9107718dc609836e6cd8abe7f2dad21..025c43e75cc204b5bb8b3bdc1083d30aaf760bbc 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
index 1e2f81915d92e26e96290f71b9b73c3465ae8130..1cc30fd9e6ad10336a49a261b3ee7c47f9053435 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
index 181ee4b428ba236d92cf71bbf33b24e53b2491c0..1e03405536a20a9f050a7943243f1c406594b836 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
@@ -1,4 +1,8 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
 #pragma once
+
 #include <iostream>
 #include <vector>
 
@@ -31,7 +35,7 @@ struct DeviceGroupedGemm : public BaseOperator
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
 
-    static_assert(DsLayout::Size() == DsDataType::Size(), "wrong! inconsisiten NumDTensor");
+    static_assert(DsLayout::Size() == DsDataType::Size(), "wrong! inconsistent NumDTensor");
 
     virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(std::vector<const void*>& p_a,
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
index 03ccc3faa5e72497017435f01f8b1b9de7d8185d..cfcafd0320efa534f44220cf501d7d6d770d5ea4 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..06d180d30fb0f3c8e3b348dd1668ff582de87493
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
@@ -0,0 +1,39 @@
+#pragma once
+#include <iostream>
+#include <vector>
+
+#include "device_grouped_gemm.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation>
+struct DeviceGroupedGemmSplitK : public DeviceGroupedGemm<ALayout,
+                                                          BLayout,
+                                                          DsLayout,
+                                                          ELayout,
+                                                          ADataType,
+                                                          BDataType,
+                                                          DsDataType,
+                                                          EDataType,
+                                                          AElementwiseOperation,
+                                                          BElementwiseOperation,
+                                                          CElementwiseOperation>
+{
+    virtual void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp b/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..bf81ed9f5b1c2e6b46208ffede15370a3ae78088
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// For pooling which used indexable operation, such as MaxPool, MinPool...etc
+template <typename DOutDataType, typename IndexDataType, typename DInDataType>
+struct DeviceIndexPoolBwd : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_dout,
+                        const void* p_indices,
+                        void* p_din,
+                        index_t dout_length,
+                        index_t din_length,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> window_strides) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp b/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
index 93202e352e8a8df349eb08f864d160e59520465b..f68022ca04de757fee1240ced4b504601b183d91 100644
--- a/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -32,8 +32,8 @@ struct DeviceMultipleReduce : public BaseOperator
         const std::array<index_t, NumOutputDim> outLengths,
         const std::array<std::array<index_t, NumOutputDim>, NumReduction> outStrides,
         const std::array<int, NumReduceDim> reduceDims,
-        const std::array<const void*, NumReduction> alphas,
-        const std::array<const void*, NumReduction> betas,
+        const std::array<double, NumReduction> alphas,
+        const std::array<double, NumReduction> betas,
         const void* in_dev,
         const std::array<void*, NumReduction> out_dev_buffers,
         const InElementwiseOperationTuple in_elementwise_op_tuple,
diff --git a/include/ck/tensor_operation/gpu/device/device_normalization.hpp b/include/ck/tensor_operation/gpu/device/device_normalization.hpp
index 227c352cbd755f9f8017cf7e3231e689447465dd..1f178f9fcb65ffdd7ab09146d94131c1d5c993f2 100644
--- a/include/ck/tensor_operation/gpu/device/device_normalization.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_normalization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,9 +14,9 @@ namespace device {
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
-          typename AccDataType,
+          typename ComputeDataType,
           typename YDataType,
-          typename AccElementwiseOperation,
+          typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim>
 struct DeviceNormalization : public BaseOperator
@@ -28,14 +28,14 @@ struct DeviceNormalization : public BaseOperator
                         const std::vector<index_t> betaStrides,
                         const std::vector<index_t> yStrides,
                         const std::vector<index_t> reduceDims,
-                        AccDataType epsilon,
+                        double epsilon,
                         const void* p_x,
                         const void* p_gamma,
                         const void* p_beta,
                         void* p_y,
                         void* p_savedMean,
                         void* p_savedInvVar,
-                        AccElementwiseOperation acc_elementwise_op) = 0;
+                        YElementwiseOperation y_elementwise_op) = 0;
 
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
 };
@@ -43,17 +43,17 @@ struct DeviceNormalization : public BaseOperator
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
-          typename AccDataType,
+          typename ComputeDataType,
           typename YDataType,
-          typename AccElementwiseOperation,
+          typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim>
 using DeviceNormalizationPtr = std::unique_ptr<DeviceNormalization<XDataType,
                                                                    GammaDataType,
                                                                    BetaDataType,
-                                                                   AccDataType,
+                                                                   ComputeDataType,
                                                                    YDataType,
-                                                                   AccElementwiseOperation,
+                                                                   YElementwiseOperation,
                                                                    Rank,
                                                                    NumReduceDim>>;
 
diff --git a/include/ck/tensor_operation/gpu/device/device_permute.hpp b/include/ck/tensor_operation/gpu/device/device_permute.hpp
index 9daa2be37338c8309bd2131f263d324fe15b7ab2..c994cf02c6aab5b85fcd2fcc29f2716de28b861d 100644
--- a/include/ck/tensor_operation/gpu/device/device_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp
deleted file mode 100644
index 3b376c6f73f066729fa118013970682ca31b340e..0000000000000000000000000000000000000000
--- a/include/ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <array>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-#include "ck/utility/reduction_enums.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-
-template <ck::ReduceTensorOp ReduceOpId>
-struct DevicePool2dFwd : public BaseOperator
-{
-    virtual std::unique_ptr<BaseArgument>
-    MakeArgumentPointer(const void* in_dev,
-                        void* out_dev,
-                        void* out_indices_dev,
-                        ck::index_t N,
-                        ck::index_t C,
-                        std::array<ck::index_t, 2> input_spatial_lengths,
-                        std::array<ck::index_t, 2> window_spatial_lengths,
-                        std::array<ck::index_t, 2> output_spatial_lengths,
-                        std::array<ck::index_t, 2> window_strides,
-                        std::array<ck::index_t, 2> input_left_pads,
-                        std::array<ck::index_t, 2> input_right_pads) = 0;
-
-    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
-};
-
-template <ck::ReduceTensorOp ReduceOpId>
-using DevicePool2dFwdPtr = std::unique_ptr<DevicePool2dFwd<ReduceOpId>>;
-
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..8b227fdfbe62798f32c98561f5235ef0c5263de4
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <index_t InOutRank,
+          index_t WindowRank,
+          typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+struct DevicePoolFwd : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in_dev,
+                        void* p_out_dev,
+                        void* p_out_indices_dev,
+                        std::vector<ck::index_t> input_lengths,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> output_lengths,
+                        std::vector<ck::index_t> input_stride,
+                        std::vector<ck::index_t> output_stride,
+                        std::vector<ck::index_t> indices_stride,
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads,
+                        std::vector<ck::index_t> pooling_dims) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_put_element.hpp b/include/ck/tensor_operation/gpu/device/device_put_element.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..9186827491c04715926fd6a981b1f7f1e626f55e
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_put_element.hpp
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// output[indices] = input
+template <typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation,
+          InMemoryDataOperationEnum Op>
+struct DevicePutElement : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_input,
+                        const void* p_indices,
+                        void* p_output,
+                        index_t input_length,
+                        index_t output_length,
+                        ElementwiseOperation elementwise_op) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_reduce.hpp b/include/ck/tensor_operation/gpu/device/device_reduce.hpp
index 15aeb8e91cdb562cc57663f6eeecd7a72668e5a2..c2721b18455c6a46e5973fad32767ae43247f59b 100644
--- a/include/ck/tensor_operation/gpu/device/device_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -13,10 +13,16 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-template <index_t Rank,
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
           index_t NumReduceDim,
+          typename ReduceOperation,
           typename InElementwiseOperation,
-          typename AccElementwiseOperation>
+          typename AccElementwiseOperation,
+          bool PropagateNan,
+          bool OutputIndex>
 struct DeviceReduce : public BaseOperator
 {
     static constexpr index_t NumOutDim = (Rank - NumReduceDim == 0) ? 1 : Rank - NumReduceDim;
@@ -27,8 +33,8 @@ struct DeviceReduce : public BaseOperator
                         const std::array<index_t, NumOutDim> outLengths,
                         const std::array<index_t, NumOutDim> outStrides,
                         const std::array<int, NumReduceDim> reduceDims,
-                        float alpha,
-                        float beta,
+                        double alpha,
+                        double beta,
                         const void* in_dev,
                         const void* in_index_dev,
                         void* out_dev,
@@ -39,12 +45,26 @@ struct DeviceReduce : public BaseOperator
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
 };
 
-template <index_t Rank,
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
           index_t NumReduceDim,
+          typename ReduceOperation,
           typename InElementwiseOperation,
-          typename AccElementwiseOperation>
-using DeviceReducePtr = std::unique_ptr<
-    DeviceReduce<Rank, NumReduceDim, InElementwiseOperation, AccElementwiseOperation>>;
+          typename AccElementwiseOperation,
+          bool PropagateNan,
+          bool OutputIndex>
+using DeviceReducePtr = std::unique_ptr<DeviceReduce<InDataType,
+                                                     AccDataType,
+                                                     OutDataType,
+                                                     Rank,
+                                                     NumReduceDim,
+                                                     ReduceOperation,
+                                                     InElementwiseOperation,
+                                                     AccElementwiseOperation,
+                                                     PropagateNan,
+                                                     OutputIndex>>;
 
 } // namespace device
 } // namespace tensor_operation
diff --git a/include/ck/tensor_operation/gpu/device/device_softmax.hpp b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
index 676e0812b740881bdd63d909b4fdf1e19aed88a3..1902fd09eec1ed01954b6cad47b0e97ff898eb40 100644
--- a/include/ck/tensor_operation/gpu/device/device_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -18,7 +18,8 @@ template <typename InDataType,
           typename OutDataType,
           typename InElementwiseOp,
           typename AccElementwiseOp,
-          index_t Rank>
+          index_t Rank,
+          index_t NumReduceDim>
 struct DeviceSoftmax : public BaseOperator
 {
     //
@@ -27,10 +28,8 @@ struct DeviceSoftmax : public BaseOperator
     // @param[in]  inLengths           Input tensor extent(s) from high to low dimension
     // @param[in]  inStrides           Input tensor stride(s) from high to low dimension
     // @param[in]  reduceDims          The dimension(s) the normalization operation is applied
-    // @param[in]  alpha               Typeless pointer in host memory storing the alpha scaling
-    //                                 value as type AccDataType
-    // @param[in]  beta                Typeless pointer in host memory storing the beta scaling
-    //                                 value as type AccDataType
+    // @param[in]  alpha               double type value
+    // @param[in]  beta                double type value
     // @param[in]  in_dev              Typeless const pointer in device memory storing the input
     //                                 tensor
     // @param      out_dev             Typeless pointer in device memory storing the output tensor
@@ -43,16 +42,14 @@ struct DeviceSoftmax : public BaseOperator
     MakeArgumentPointer(const std::vector<index_t> inLengths,
                         const std::vector<index_t> inStrides,
                         const std::vector<int> reduceDims,
-                        const void* alpha,
-                        const void* beta,
+                        double alpha,
+                        double beta,
                         const void* in_dev,
                         void* out_dev,
                         InElementwiseOp in_elementwise_op,
                         AccElementwiseOp acc_elementwise_op) = 0;
 
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
-    virtual index_t GetRank() const                           = 0;
-    virtual index_t GetNumReduceDim() const                   = 0;
 };
 
 template <typename InDataType,
@@ -60,9 +57,15 @@ template <typename InDataType,
           typename OutDataType,
           typename InElementwiseOp,
           typename AccElementwiseOp,
-          index_t Rank>
-using DeviceSoftmaxPtr = std::unique_ptr<
-    DeviceSoftmax<InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank>>;
+          index_t Rank,
+          index_t NumReduceDim>
+using DeviceSoftmaxPtr = std::unique_ptr<DeviceSoftmax<InDataType,
+                                                       AccDataType,
+                                                       OutDataType,
+                                                       InElementwiseOp,
+                                                       AccElementwiseOp,
+                                                       Rank,
+                                                       NumReduceDim>>;
 
 } // namespace device
 } // namespace tensor_operation
diff --git a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
index f59e6093e2ae024ccbf8827082c755eba8cee88f..eeccd977ccbf4440a349845826d45d5a3274ad37 100644
--- a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp b/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
index fc913e9ba03611dfbabea020bd4054c0b953726d..0bb45b18c3e19b2ec5f9347c1e811d8734ee45a9 100644
--- a/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..4d599e8017991b4f1b1905de9acd23a219ac67f9
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
@@ -0,0 +1,992 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// Tensor Contraction:
+//   input : A
+//   input : B
+//   input : D0, D1, ...
+//   output : E
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   A[G0, G1, ..., M0, M1, M2, ..., K0, K1, K2, ...]
+//   B[G0, G1, ..., N0, N1, N2, ..., K0, K1, K2, ...]
+//   D[G0, G1, ..., M0, M1, M2, ..., N0, N1, N2, ...]
+//   E[G0, G1, ..., M0, M1, M2, ..., N0, N1, N2, ...]
+
+// NOTE: TensorSpecialization::Packed specialized tensor is "packed" in a sense that each inner
+// dimension in a dimension group (eg [G0, G1] in Gs, [M0, M1, M2] in Ms, etc.) are contiguous and
+// ordered. Not in a sense that the tensor [G0, G1, ..., M0, M1, ..., N0, N1...] can be permuted
+// while still being a contiguous, unpadded tensor. In other words, it merely degenerates into
+// TensorSpecialization::Default with NumDimG/M/N/K = 1
+//
+// Detail- Packed tensor satisfies
+//   stride_0 = 1
+//   stride_i = stride_{i - 1} * extent_{i - 1}
+// So tensor
+//   [G0, G1, G2, M, N]
+// transposed into tensor
+//   [G0, G2, G1, M, N]
+// with strides
+//   [G2 * G1 * M * N, G1 * M * N, M * N, N, 1]
+// is again a packed tensor. MakeGridDescriptor() currently just merges dimensions and ignores some
+// strides from input tensor extents so finer dimension information is lost. Merging dimensions is
+// essentially a degenerated case of TensorSpecialization::Default with NumDimG/M/N/K = 1.
+//
+// Might need to expose dimension order to the interface to fully support
+// TensorSpecialization::Packed in a traditional sense of "packed" tensor
+template <index_t NumDimG,
+          index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          TensorSpecialization ASpec,
+          TensorSpecialization BSpec,
+          TensorSpecialization DESpec,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t K0PerBlock,
+          ck::index_t K1,
+          ck::index_t MPerWMMA,
+          ck::index_t NPerWMMA,
+          ck::index_t MRepeat,
+          ck::index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsAddExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsAddExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+          ck::index_t NumPrefetch         = 1,
+          ck::LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
+struct DeviceBatchedContractionMultipleD_Wmma_CShuffle
+    : public DeviceBatchedContractionMultipleD<NumDimG,
+                                               NumDimM,
+                                               NumDimN,
+                                               NumDimK,
+                                               ADataType,
+                                               BDataType,
+                                               DsDataType,
+                                               EDataType,
+                                               AElementwiseOperation,
+                                               BElementwiseOperation,
+                                               CDEElementwiseOperation>
+{
+    using DeviceOp                      = DeviceBatchedContractionMultipleD_Wmma_CShuffle;
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    // K1 = Max Vector Access Pixels
+    static constexpr auto K1Number = Number<K1>{};
+
+    static constexpr auto matrix_padder =
+        MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, K0PerBlock* K1};
+
+    // Assume: A[G0, G1, ..., M0, M1, M2, ..., K0, K1, K2, ...]
+    static auto MakeAGridDescriptor_M_K(const std::vector<index_t>& a_gs_ms_ks_lengths_vec,
+                                        const std::vector<index_t>& a_gs_ms_ks_strides_vec)
+    {
+        assert(a_gs_ms_ks_lengths_vec.size() == NumDimG + NumDimM + NumDimK &&
+               a_gs_ms_ks_strides_vec.size() == NumDimG + NumDimM + NumDimK);
+
+        const auto to_tuple = [&](auto& vec, auto start, auto end) {
+            return generate_tuple([&](auto i) { return vec[start + i]; }, Number<end - start>{});
+        };
+
+        const auto a_ms_ks_lengths = to_tuple(
+            a_gs_ms_ks_lengths_vec, Number<NumDimG>{}, Number<NumDimG + NumDimM + NumDimK>{});
+        const auto a_ms_ks_strides = to_tuple(
+            a_gs_ms_ks_strides_vec, Number<NumDimG>{}, Number<NumDimG + NumDimM + NumDimK>{});
+
+        // dimension Ids for M0, M1, ...
+        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
+
+        // dimension Ids for K0, K1, ...
+        constexpr auto kDimIds =
+            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimK, 1>::type{};
+
+        // lengths for M0, M1, ...
+        const auto mLengths = get_container_subset(a_ms_ks_lengths, mDimIds);
+
+        // lengths for K0, K1, ...
+        const auto kLengths = get_container_subset(a_ms_ks_lengths, kDimIds);
+
+        if constexpr(ASpec == TensorSpecialization::Packed)
+        {
+            auto M = container_reduce(mLengths, math::multiplies{}, Number<1>{});
+            auto K = container_reduce(kLengths, math::multiplies{}, Number<1>{});
+            const auto a_grid_desc_mraw_kraw = make_naive_tensor_descriptor(
+                make_tuple(M, K),
+                make_tuple(a_ms_ks_strides[Number<NumDimM - 1>{}],
+                           a_ms_ks_strides[Number<NumDimM + NumDimK - 1>{}]));
+            return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+        }
+        else
+        {
+            // naive tensor A[M0, M1, M2, ..., K0, K1, K2...]
+            const auto a_grid_desc_ms_ks =
+                make_naive_tensor_descriptor(a_ms_ks_lengths, a_ms_ks_strides);
+
+            // transformed tensor A[MRaw = M0 * M1 * M2 * ... , KRaw = K0 * K1 * K2 * ...]
+            const auto a_grid_desc_mraw_kraw = transform_tensor_descriptor(
+                a_grid_desc_ms_ks,
+                make_tuple(make_merge_transform(mLengths), make_merge_transform(kLengths)),
+                make_tuple(mDimIds, kDimIds),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+        }
+    }
+
+    // Assume: B[G0, G1, ..., N0, N1, N2, ..., K0, K1, K2, ...]
+    static auto MakeBGridDescriptor_N_K(const std::vector<index_t>& b_gs_ns_ks_lengths_vec,
+                                        const std::vector<index_t>& b_gs_ns_ks_strides_vec)
+    {
+        assert(b_gs_ns_ks_lengths_vec.size() == NumDimG + NumDimN + NumDimK &&
+               b_gs_ns_ks_strides_vec.size() == NumDimG + NumDimN + NumDimK);
+
+        const auto to_tuple = [&](auto& vec, auto start, auto end) {
+            return generate_tuple([&](auto i) { return vec[start + i]; }, Number<end - start>{});
+        };
+
+        const auto b_ns_ks_lengths = to_tuple(
+            b_gs_ns_ks_lengths_vec, Number<NumDimG>{}, Number<NumDimG + NumDimN + NumDimK>{});
+        const auto b_ns_ks_strides = to_tuple(
+            b_gs_ns_ks_strides_vec, Number<NumDimG>{}, Number<NumDimG + NumDimN + NumDimK>{});
+
+        // dimension Ids for N0, N1, ...
+        constexpr auto nDimIds = typename arithmetic_sequence_gen<0, NumDimN, 1>::type{};
+
+        // dimension Ids for K0, K1, ...
+        constexpr auto kDimIds =
+            typename arithmetic_sequence_gen<NumDimN, NumDimN + NumDimK, 1>::type{};
+
+        // lengths for K0, K1, ...
+        const auto kLengths = get_container_subset(b_ns_ks_lengths, kDimIds);
+
+        // lengths for N0, N1, ...
+        const auto nLengths = get_container_subset(b_ns_ks_lengths, nDimIds);
+
+        if constexpr(BSpec == TensorSpecialization::Packed)
+        {
+            auto N = container_reduce(nLengths, math::multiplies{}, Number<1>{});
+            auto K = container_reduce(kLengths, math::multiplies{}, Number<1>{});
+            const auto b_grid_desc_nraw_kraw = make_naive_tensor_descriptor(
+                make_tuple(N, K),
+                make_tuple(b_ns_ks_strides[Number<NumDimN - 1>{}],
+                           b_ns_ks_strides[Number<NumDimN + NumDimK - 1>{}]));
+            return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+        }
+        else
+        {
+            // naive tensor B[N0, N1, N2, ..., K0, K1, K2, ...]
+            const auto b_grid_desc_ns_ks =
+                make_naive_tensor_descriptor(b_ns_ks_lengths, b_ns_ks_strides);
+
+            // transformed tensor B[NRaw = N0 * N1 * N2 * ..., KRaw = K0 * K1 * K2 * ...]
+            const auto b_grid_desc_nraw_kraw = transform_tensor_descriptor(
+                b_grid_desc_ns_ks,
+                make_tuple(make_merge_transform(nLengths), make_merge_transform(kLengths)),
+                make_tuple(nDimIds, kDimIds),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+        }
+    }
+
+    // assume E[G0, G1, ..., M0, M1, M2, ..., N0, N1, N2...]
+    static auto MakeEGridDescriptor_M_N(const std::vector<index_t>& e_gs_ms_ns_lengths_vec,
+                                        const std::vector<index_t>& e_gs_ms_ns_strides_vec)
+    {
+        assert(e_gs_ms_ns_lengths_vec.size() == NumDimG + NumDimM + NumDimN &&
+               e_gs_ms_ns_strides_vec.size() == NumDimG + NumDimM + NumDimN);
+
+        const auto to_tuple = [&](auto& vec, auto start, auto end) {
+            return generate_tuple([&](auto i) { return vec[start + i]; }, Number<end - start>{});
+        };
+
+        const auto e_ms_ns_lengths = to_tuple(
+            e_gs_ms_ns_lengths_vec, Number<NumDimG>{}, Number<NumDimG + NumDimM + NumDimN>{});
+        const auto e_ms_ns_strides = to_tuple(
+            e_gs_ms_ns_strides_vec, Number<NumDimG>{}, Number<NumDimG + NumDimM + NumDimN>{});
+
+        // dimension Ids for M0, M1, ...
+        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
+
+        // dimension Ids for N0, N1, ...
+        constexpr auto nDimIds =
+            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimN, 1>::type{};
+
+        // lengths for M0, M1, ...
+        const auto mLengths = get_container_subset(e_ms_ns_lengths, mDimIds);
+
+        // lengths for K0, K1, ...
+        const auto nLengths = get_container_subset(e_ms_ns_lengths, nDimIds);
+
+        if constexpr(DESpec == TensorSpecialization::Packed)
+        {
+            auto M = container_reduce(mLengths, math::multiplies{}, Number<1>{});
+            auto N = container_reduce(nLengths, math::multiplies{}, Number<1>{});
+            const auto e_grid_desc_mraw_nraw = make_naive_tensor_descriptor(
+                make_tuple(M, N),
+                make_tuple(e_ms_ns_strides[Number<NumDimM - 1>{}],
+                           e_ms_ns_strides[Number<NumDimM + NumDimN - 1>{}]));
+            return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+        }
+        else
+        {
+            // naive tensor E[M0, M1, M2, ..., N0, N1, N2...]
+            const auto e_grid_desc_ms_ns =
+                make_naive_tensor_descriptor(e_ms_ns_lengths, e_ms_ns_strides);
+
+            // transformed tensor E[MRaw = M0 * M1 * M2 * ... , NRaw = N0 * N1 * N2 * ...]
+            const auto e_grid_desc_mraw_nraw = transform_tensor_descriptor(
+                e_grid_desc_ms_ns,
+                make_tuple(make_merge_transform(mLengths), make_merge_transform(nLengths)),
+                make_tuple(mDimIds, nDimIds),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+        }
+    }
+
+    // assume E[G0, G1, ..., M0, M1, M2, ..., N0, N1, N2...]
+    static auto MakeEGridDescriptor_G_M_N(const std::vector<index_t>& e_gs_ms_ns_lengths_vec,
+                                          const std::vector<index_t>& e_gs_ms_ns_strides_vec)
+    {
+        assert(e_gs_ms_ns_lengths_vec.size() == NumDimG + NumDimM + NumDimN &&
+               e_gs_ms_ns_strides_vec.size() == NumDimG + NumDimM + NumDimN);
+
+        const auto to_tuple = [&](auto& vec, auto start, auto end) {
+            return generate_tuple([&](auto i) { return vec[start + i]; }, Number<end - start>{});
+        };
+
+        const auto e_gs_ms_ns_lengths =
+            to_tuple(e_gs_ms_ns_lengths_vec, Number<0>{}, Number<NumDimG + NumDimM + NumDimN>{});
+        const auto e_gs_ms_ns_strides =
+            to_tuple(e_gs_ms_ns_strides_vec, Number<0>{}, Number<NumDimG + NumDimM + NumDimN>{});
+
+        // dimension Ids for G0, G1, ...
+        constexpr auto gDimIds = typename arithmetic_sequence_gen<0, NumDimG, 1>::type{};
+
+        // dimension Ids for M0, M1, ...
+        constexpr auto mDimIds =
+            typename arithmetic_sequence_gen<NumDimG, NumDimG + NumDimM, 1>::type{};
+
+        // dimension Ids for N0, N1, ...
+        constexpr auto nDimIds = typename arithmetic_sequence_gen<NumDimG + NumDimM,
+                                                                  NumDimG + NumDimM + NumDimN,
+                                                                  1>::type{};
+
+        // lengths for G0, G1, ...
+        const auto gLengths = get_container_subset(e_gs_ms_ns_lengths, gDimIds);
+
+        // lengths for M0, M1, ...
+        const auto mLengths = get_container_subset(e_gs_ms_ns_lengths, mDimIds);
+
+        // lengths for K0, K1, ...
+        const auto nLengths = get_container_subset(e_gs_ms_ns_lengths, nDimIds);
+
+        if constexpr(DESpec == TensorSpecialization::Packed)
+        {
+            auto G = container_reduce(gLengths, math::multiplies{}, Number<1>{});
+            auto M = container_reduce(mLengths, math::multiplies{}, Number<1>{});
+            auto N = container_reduce(nLengths, math::multiplies{}, Number<1>{});
+            const auto e_grid_desc_g_mraw_nraw = make_naive_tensor_descriptor(
+                make_tuple(G, M, N),
+                make_tuple(e_gs_ms_ns_strides[Number<NumDimG - 1>{}],
+                           e_gs_ms_ns_strides[Number<NumDimG + NumDimM - 1>{}],
+                           e_gs_ms_ns_strides[Number<NumDimG + NumDimM + NumDimN - 1>{}]));
+            // return matrix_padder.PadCDescriptor_M_N(e_grid_desc_g_mraw_nraw);
+            return e_grid_desc_g_mraw_nraw;
+        }
+        else
+        {
+            // naive tensor E[G0, G1, ..., M0, M1, M2, ..., N0, N1, N2...]
+            const auto e_grid_desc_gs_ms_ns =
+                make_naive_tensor_descriptor(e_gs_ms_ns_lengths, e_gs_ms_ns_strides);
+
+            // transformed tensor E[G = G0 * G1 * ..., MRaw = M0 * M1 * M2 * ... , NRaw = N0 * N1 *
+            // N2 * ...]
+            const auto e_grid_desc_g_mraw_nraw = transform_tensor_descriptor(
+                e_grid_desc_gs_ms_ns,
+                make_tuple(make_merge_transform(gLengths),
+                           make_merge_transform(mLengths),
+                           make_merge_transform(nLengths)),
+                make_tuple(gDimIds, mDimIds, nDimIds),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+            // return matrix_padder.PadCDescriptor_M_N(e_grid_desc_g_mraw_nraw);
+            return e_grid_desc_g_mraw_nraw;
+        }
+    }
+
+    static auto MakeDsGridDescriptor_M_N(
+        const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_lengths_vec,
+        const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_strides_vec)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return DeviceOp::MakeEGridDescriptor_M_N(ds_gs_ms_ns_lengths_vec[i],
+                                                         ds_gs_ms_ns_strides_vec[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    static auto MakeDsGridDescriptor_G_M_N(
+        const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_lengths_vec,
+        const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_strides_vec)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return DeviceOp::MakeEGridDescriptor_G_M_N(ds_gs_ms_ns_lengths_vec[i],
+                                                           ds_gs_ms_ns_strides_vec[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // Gridwise descriptor, mapping to whole given provblem.
+    using AGridDesc_M_K  = decltype(MakeAGridDescriptor_M_K({}, {}));
+    using BGridDesc_N_K  = decltype(MakeBGridDescriptor_N_K({}, {}));
+    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
+    using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N({}, {}));
+
+    using DsGridDesc_G_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_G_M_N({}, {}))>;
+    using EGridDesc_G_M_N  = decltype(MakeEGridDescriptor_G_M_N({}, {}));
+
+    struct ComputePtrOffsetOfStridedBatch
+    {
+        ComputePtrOffsetOfStridedBatch(index_t batch_stride_A,
+                                       index_t batch_stride_B,
+                                       DsGridDesc_G_M_N ds_grid_desc_g_m_n,
+                                       EGridDesc_G_M_N e_grid_desc_g_m_n)
+            : batch_stride_A_(batch_stride_A),
+              batch_stride_B_(batch_stride_B),
+              ds_grid_desc_g_m_n_(ds_grid_desc_g_m_n),
+              e_grid_desc_g_m_n_(e_grid_desc_g_m_n)
+        {
+        }
+
+        __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
+        {
+            return static_cast<long_index_t>(g_idx) * batch_stride_A_;
+        }
+
+        __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
+        {
+            return static_cast<long_index_t>(g_idx) * batch_stride_B_;
+        }
+
+        __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
+        {
+            std::array<long_index_t, NumDTensor> ds_offset;
+
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                ds_offset[i] = static_cast<long_index_t>(g_idx) *
+                               ds_grid_desc_g_m_n_[i].CalculateOffset(make_multi_index(1, 0, 0));
+            });
+
+            return ds_offset;
+        }
+
+        __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
+        {
+            return static_cast<long_index_t>(g_idx) *
+                   e_grid_desc_g_m_n_.CalculateOffset(make_multi_index(1, 0, 0));
+        }
+
+        private:
+        index_t batch_stride_A_;
+        index_t batch_stride_B_;
+        DsGridDesc_G_M_N ds_grid_desc_g_m_n_;
+        EGridDesc_G_M_N e_grid_desc_g_m_n_;
+    };
+
+    // A desc for source in blockwise copy
+    template <typename AGridDesc_M_K>
+    __host__ __device__ static constexpr auto
+    MakeAGridDescriptor_K0_M_K1(const AGridDesc_M_K& a_grid_desc_m_k)
+    {
+        const auto M = a_grid_desc_m_k.GetLength(I0);
+        const auto K = a_grid_desc_m_k.GetLength(I1);
+
+        const auto AK0 = K / K1;
+
+        return transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_unmerge_transform(make_tuple(AK0, K1)), make_pass_through_transform(M)),
+            make_tuple(Sequence<1>{}, Sequence<0>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    // B desc for source in blockwise copy
+    template <typename BGridDesc_N_K>
+    __host__ __device__ static constexpr auto
+    MakeBGridDescriptor_K0_N_K1(const BGridDesc_N_K& b_grid_desc_n_k)
+    {
+        const auto N = b_grid_desc_n_k.GetLength(I0);
+        const auto K = b_grid_desc_n_k.GetLength(I1);
+
+        const auto BK0 = K / K1;
+
+        return transform_tensor_descriptor(
+            b_grid_desc_n_k,
+            make_tuple(make_unmerge_transform(make_tuple(BK0, K1)), make_pass_through_transform(N)),
+            make_tuple(Sequence<1>{}, Sequence<0>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    using AGridDesc_K0_M_K1 = decltype(DeviceOp::MakeAGridDescriptor_K0_M_K1(AGridDesc_M_K{}));
+    using BGridDesc_K0_N_K1 = decltype(DeviceOp::MakeBGridDescriptor_K0_N_K1(BGridDesc_N_K{}));
+
+    // GridwiseOp
+    using GridwiseOp = GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle<
+        // DataType Family
+        ADataType,
+        BDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        // InMemory Data Descriptor
+        AGridDesc_K0_M_K1,
+        BGridDesc_K0_N_K1,
+        DsGridDesc_M_N,
+        EGridDesc_M_N,
+        // ElementwiseOp Family
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        // Tiling Family
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerWMMA,
+        NPerWMMA,
+        K1,
+        MRepeat,
+        NRepeat,
+        // ThreadCluster Family
+        BlockSize,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+        NumPrefetch,
+        LoopSched,
+        PipelineVer>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a_grid,
+                 const void* p_b_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 const std::vector<index_t>& a_gs_ms_ks_lengths,
+                 const std::vector<index_t>& b_gs_ns_ks_lengths,
+                 const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_lengths,
+                 const std::vector<index_t>& e_gs_ms_ns_lengths,
+                 const std::vector<index_t>& a_gs_ms_ks_strides,
+                 const std::vector<index_t>& b_gs_ns_ks_strides,
+                 const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_strides,
+                 const std::vector<index_t>& e_gs_ms_ns_strides,
+                 index_t M01,
+                 index_t N01,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
+              p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              a_grid_desc_m_k_{},
+              b_grid_desc_n_k_{},
+              ds_grid_desc_m_n_{},
+              e_grid_desc_m_n_{},
+              ds_grid_desc_g_m_n_{
+                  DeviceOp::MakeDsGridDescriptor_G_M_N(ds_gs_ms_ns_lengths, ds_gs_ms_ns_strides)},
+              e_grid_desc_g_m_n_{
+                  DeviceOp::MakeEGridDescriptor_G_M_N(e_gs_ms_ns_lengths, e_gs_ms_ns_strides)},
+              a_grid_desc_k0_m_k1_{},
+              b_grid_desc_k0_n_k1_{},
+              ds_grid_desc_mblock_mperblock_nblock_nperblock{},
+              e_grid_desc_mblock_mperblock_nblock_nperblock{},
+              block_2_ctile_map_{},
+              M01_{M01},
+              N01_{N01},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op},
+              a_mz_stride_{},
+              a_kz_stride_{},
+              b_nz_stride_{},
+              b_kz_stride_{},
+              ds_nz_stride_{},
+              e_nz_stride_{},
+              a_batch_stride_{a_gs_ms_ks_strides[NumDimG - 1]},
+              b_batch_stride_{b_gs_ns_ks_strides[NumDimG - 1]},
+              compute_ptr_offset_of_batch_{
+                  a_batch_stride_, b_batch_stride_, ds_grid_desc_g_m_n_, e_grid_desc_g_m_n_}
+        {
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+            });
+
+            a_grid_desc_m_k_ =
+                DeviceOp::MakeAGridDescriptor_M_K(a_gs_ms_ks_lengths, a_gs_ms_ks_strides);
+            b_grid_desc_n_k_ =
+                DeviceOp::MakeBGridDescriptor_N_K(b_gs_ns_ks_lengths, b_gs_ns_ks_strides);
+
+            ds_grid_desc_m_n_ =
+                DeviceOp::MakeDsGridDescriptor_M_N(ds_gs_ms_ns_lengths, ds_gs_ms_ns_strides);
+
+            e_grid_desc_m_n_ =
+                DeviceOp::MakeEGridDescriptor_M_N(e_gs_ms_ns_lengths, e_gs_ms_ns_strides);
+
+            a_grid_desc_k0_m_k1_ = DeviceOp::MakeAGridDescriptor_K0_M_K1(a_grid_desc_m_k_);
+            b_grid_desc_k0_n_k1_ = DeviceOp::MakeBGridDescriptor_K0_N_K1(b_grid_desc_n_k_);
+
+            block_2_ctile_map_ = GridwiseOp::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_, M01, N01);
+
+            ds_grid_desc_mblock_mperblock_nblock_nperblock =
+                GridwiseOp::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                    ds_grid_desc_m_n_);
+
+            e_grid_desc_mblock_mperblock_nblock_nperblock =
+                GridwiseOp::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(e_grid_desc_m_n_);
+
+            // for sanity check of vector memory access
+            a_mz_stride_ = a_gs_ms_ks_strides[NumDimG + NumDimM - 1];
+            a_kz_stride_ = a_gs_ms_ks_strides[NumDimG + NumDimM + NumDimK - 1];
+            b_nz_stride_ = b_gs_ns_ks_strides[NumDimG + NumDimN - 1];
+            b_kz_stride_ = b_gs_ns_ks_strides[NumDimG + NumDimN + NumDimK - 1];
+
+            for(index_t i = 0; i < NumDTensor; ++i)
+            {
+                ds_nz_stride_[i] = ds_gs_ms_ns_strides[i][NumDimG + NumDimM + NumDimN - 1];
+            }
+
+            e_nz_stride_ = e_gs_ms_ns_strides[NumDimG + NumDimM + NumDimN - 1];
+        }
+
+        // Pointers
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseOp::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        // Tensor Descriptors
+        AGridDesc_M_K a_grid_desc_m_k_;
+        BGridDesc_N_K b_grid_desc_n_k_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+        DsGridDesc_G_M_N ds_grid_desc_g_m_n_;
+        EGridDesc_G_M_N e_grid_desc_g_m_n_;
+
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+
+        typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            ds_grid_desc_mblock_mperblock_nblock_nperblock;
+        typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            e_grid_desc_mblock_mperblock_nblock_nperblock;
+
+        // Block to Tile mapping
+        typename GridwiseOp::DefaultBlock2CTileMap block_2_ctile_map_;
+
+        // Idle
+        index_t M01_;
+        index_t N01_;
+
+        // ElementwiseOp
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+
+        // Strides for the last M/N/K dimensions of A/B/Ds/E
+        //   for sanity check of vector load/store
+        index_t a_mz_stride_;
+        index_t a_kz_stride_;
+        index_t b_nz_stride_;
+        index_t b_kz_stride_;
+        std::array<index_t, NumDTensor> ds_nz_stride_;
+        index_t e_mz_stride_;
+        index_t e_nz_stride_;
+
+        index_t a_batch_stride_;
+        index_t b_batch_stride_;
+
+        // Batch Offset
+        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            const index_t G = arg.e_grid_desc_g_m_n_.GetLength(I0);
+
+            const index_t grid_size =
+                arg.block_2_ctile_map_.CalculateGridSize(arg.e_grid_desc_m_n_) * G;
+
+            const auto K =
+                arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel = kernel_contraction_multiple_d_wmma_cshuffle<
+                    GridwiseOp,
+                    ADataType,
+                    BDataType,
+                    typename GridwiseOp::DsGridPointer,
+                    EDataType,
+                    DeviceOp::AGridDesc_K0_M_K1,
+                    DeviceOp::BGridDesc_K0_N_K1,
+                    typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                    typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CDEElementwiseOperation,
+                    ComputePtrOffsetOfStridedBatch,
+                    typename GridwiseOp::DefaultBlock2CTileMap,
+                    has_main_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_a_grid_,
+                                              arg.p_b_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              G,
+                                              arg.a_grid_desc_k0_m_k1_,
+                                              arg.b_grid_desc_k0_n_k1_,
+                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.compute_ptr_offset_of_batch_,
+                                              arg.block_2_ctile_map_);
+            };
+
+            if(GridwiseOp::CalculateHasMainKBlockLoop(K))
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
+        {
+            if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        if(!GridwiseOp::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                      arg.b_grid_desc_k0_n_k1_,
+                                      arg.ds_grid_desc_m_n_,
+                                      arg.e_grid_desc_m_n_,
+                                      arg.block_2_ctile_map_))
+        {
+            return false;
+        }
+
+        // check vector access
+        static_assert((ABlockTransferSrcVectorDim == 1 || ABlockTransferSrcVectorDim == 2) &&
+                          (BBlockTransferSrcVectorDim == 1 || BBlockTransferSrcVectorDim == 2),
+                      "wrong!");
+
+        // vector memory access of A: could be on M or AK1 dimension
+        if constexpr(ABlockTransferSrcVectorDim == 1)
+        {
+            if(!(arg.a_mz_stride_ == 1 &&
+                 arg.a_grid_desc_k0_m_k1_.GetLength(I1) % ABlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(!(arg.a_kz_stride_ == 1 &&
+                 arg.a_grid_desc_k0_m_k1_.GetLength(I2) % ABlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+
+        // vector memory access of B: could be on N or BK1 dimension
+        if constexpr(BBlockTransferSrcVectorDim == 1)
+        {
+            if(!(arg.b_nz_stride_ == 1 &&
+                 arg.b_grid_desc_k0_n_k1_.GetLength(I1) % BBlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(!(arg.b_kz_stride_ == 1 &&
+                 arg.b_grid_desc_k0_n_k1_.GetLength(I2) % BBlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+
+        // vector memory access of Ds: always on NPerBlock dimension
+        bool valid_d_access = true;
+
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            if(!(arg.ds_nz_stride_[i] == 1 &&
+                 arg.ds_grid_desc_mblock_mperblock_nblock_nperblock[i].GetLength(I3) %
+                         CDEShuffleBlockTransferScalarPerVector_NPerBlock ==
+                     0))
+            {
+                valid_d_access = false;
+            }
+        });
+
+        if(valid_d_access == false)
+        {
+            return false;
+        }
+
+        // vector memory access of E: always on NPerBlock dimension
+        if(!((arg.e_nz_stride_ == 1 &&
+              arg.e_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I3) %
+                      CDEShuffleBlockTransferScalarPerVector_NPerBlock ==
+                  0) ||
+             CDEShuffleBlockTransferScalarPerVector_NPerBlock == 1))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto
+    MakeArgument(const void* p_a,
+                 const void* p_b,
+                 std::array<const void*, NumDTensor> p_ds,
+                 void* p_e,
+                 const std::vector<index_t>& a_gs_ms_ks_lengths,
+                 const std::vector<index_t>& a_gs_ms_ks_strides,
+                 const std::vector<index_t>& b_gs_ns_ks_lengths,
+                 const std::vector<index_t>& b_gs_ns_ks_strides,
+                 const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_lengths,
+                 const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_strides,
+                 const std::vector<index_t>& e_gs_ms_ns_lengths,
+                 const std::vector<index_t>& e_gs_ms_ns_strides,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_e,
+                        a_gs_ms_ks_lengths,
+                        b_gs_ns_ks_lengths,
+                        ds_gs_ms_ns_lengths,
+                        e_gs_ms_ns_lengths,
+                        a_gs_ms_ks_strides,
+                        b_gs_ns_ks_strides,
+                        ds_gs_ms_ns_strides,
+                        e_gs_ms_ns_strides,
+                        1,
+                        1,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const std::vector<index_t>& a_gs_ms_ks_lengths,
+                        const std::vector<index_t>& a_gs_ms_ks_strides,
+                        const std::vector<index_t>& b_gs_ns_ks_lengths,
+                        const std::vector<index_t>& b_gs_ns_ks_strides,
+                        const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_lengths,
+                        const std::array<std::vector<index_t>, NumDTensor>& ds_gs_ms_ns_strides,
+                        const std::vector<index_t>& e_gs_ms_ns_lengths,
+                        const std::vector<index_t>& e_gs_ms_ns_strides,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          a_gs_ms_ks_lengths,
+                                          b_gs_ns_ks_lengths,
+                                          ds_gs_ms_ns_lengths,
+                                          e_gs_ms_ns_lengths,
+                                          a_gs_ms_ks_strides,
+                                          b_gs_ns_ks_strides,
+                                          ds_gs_ms_ns_strides,
+                                          e_gs_ms_ns_strides,
+                                          1,
+                                          1,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<LoopScheduler, std::string> LoopSchedToString{
+            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceBatchedContractionMultipleD_Wmma_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerWMMA << ", "
+            << NPerWMMA << ", "
+            << MRepeat << ", "
+            << NRepeat
+            << ">"
+            << " NumPrefetch: "
+            << NumPrefetch << ", "
+            << "LoopScheduler: "
+            << LoopSchedToString[LoopSched] << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
index 2237ad944cd61775267851a9a2629efb115f286f..46e71240c1f5a655e3f6c2b95b1dc9d10214b6f1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -56,7 +56,8 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t num_blocks_per_batch =
@@ -839,7 +840,9 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
index 01f5e17d91436747d0b3bf052061741b34e0b46a..fc080df5f5d096a7d5b2166da28786bb977127e7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
@@ -74,7 +74,8 @@ __global__ void
                                           const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
                                           const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
index 3b87e56337f95e2463a8211c961504fceb764d8b..7012584aabfb50f79408ccb663f4cd427d1ccfe2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -60,7 +60,8 @@ __global__ void
             const index_t batch_count,
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -588,7 +589,9 @@ struct DeviceBatchedGemmGemm_Xdl_CShuffle : public DeviceBatchedGemmGemm<ALayout
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
index e34c19bdf02841d0f9c0b1a381570f7e34120b60..e00deaf000bc4526861f9e3c9cfef9f609796211 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -83,7 +83,8 @@ __global__ void
                                 const Block2ETileMap block_2_etile_map)
 {
 
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -579,7 +580,9 @@ struct DeviceBatchedGemmMultiD_Xdl : public DeviceBatchedGemmMultiD<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0df365ceb29c2892d5028428816988a73a43f322
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
@@ -0,0 +1,796 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+/*
+ * \tparam ComputePtrOffsetOfBatch Class that computes the base pointer offsets of A, B, C matrix
+ * given the batch. For example, ComputePtrOffsetOfStridedBatch() computes the offsets of evenly
+ * strided batched, but we can easily extend to other layouts. The returned offset can be either \p
+ * index_t or \p long_index_t. If it returns \p long_index_t, we are not subject to the 2GB
+ * limitations.
+ *
+ * \note Using \p ComputePtrOffsetOfBatch gives us the flexibility that 2 workgroups can compute 2
+ * tiles from different matrices. Keep in mind that these 2 matrices can share the same grid
+ * descriptor (like in BatchedGEMM), or use their own grid descriptors (in GroupedGemm). \link
+ * impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp kernel_gemm_xdlops_v2r3_for_conv3d \endlink for
+ * \link DeviceConv3d \endlink uses the same concept, but currently does NOT encapsulate the
+ * computing of pointer offset into \p ComputePtrOffsetOfStridedBatch.
+ */
+
+template <typename GridwiseGemm,
+          typename ABDataType,
+          typename DsPointer,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AGridDesc_K0_M0_M1_K1,
+          typename BGridDesc_K0_N0_N1_K1,
+          typename DsGridDesc_M0_M10_M11_N0_N10_N11,
+          typename CGridDesc_M0_M10_M11_N0_N10_N11,
+          typename ComputePtrOffsetOfBatch,
+          typename Block2CTileMap,
+          bool HasMainKBlockLoop,
+          bool HasDoubleTailKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_dl_multiple_d(
+            const ABDataType* __restrict__ p_a_grid,
+            const ABDataType* __restrict__ p_b_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const index_t batch_count,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const AGridDesc_K0_M0_M1_K1 a_grid_desc_k0_m0_m1_k1,
+            const BGridDesc_K0_N0_N1_K1 b_grid_desc_k0_n0_n1_k1,
+            const DsGridDesc_M0_M10_M11_N0_N10_N11 ds_grid_desc_m0_m10_m11_n0_n10_n11,
+            const CGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11,
+            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
+            const Block2CTileMap block_2_ctile_map)
+{
+// TODO: Enable for gfx90a after complier fix
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||              \
+    defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__))
+
+    const index_t num_blocks_per_batch =
+        __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+    const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
+
+    const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
+    const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
+    const long_index_t e_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetEPtrOffset(g_idx)));
+
+    const auto ds_batch_offset = compute_ptr_offset_of_batch.GetDsPtrOffset(g_idx);
+
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    DsPointer p_ds_grid_grp;
+
+    static constexpr index_t NumDTensor = DsGridDesc_M0_M10_M11_N0_N10_N11::Size();
+
+    static_for<0, NumDTensor, 1>{}(
+        [&](auto i) { p_ds_grid_grp(i) = p_ds_grid[i] + ds_batch_offset[i]; });
+
+    GridwiseGemm::Run(p_a_grid + a_batch_offset,
+                      p_b_grid + b_batch_offset,
+                      p_ds_grid_grp,
+                      p_e_grid + e_batch_offset,
+                      p_shared,
+                      a_element_op,
+                      b_element_op,
+                      cde_element_op,
+                      a_grid_desc_k0_m0_m1_k1,
+                      b_grid_desc_k0_n0_n1_k1,
+                      ds_grid_desc_m0_m10_m11_n0_n10_n11,
+                      e_grid_desc_m0_m10_m11_n0_n10_n11,
+                      block_2_ctile_map,
+                      integral_constant<bool, HasMainKBlockLoop>{},
+                      integral_constant<bool, HasDoubleTailKBlockLoop>{});
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = batch_count;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = a_grid_desc_k0_m0_m1_k1;
+    ignore = b_grid_desc_k0_n0_n1_k1;
+    ignore = ds_grid_desc_m0_m10_m11_n0_n10_n11;
+    ignore = e_grid_desc_m0_m10_m11_n0_n10_n11;
+    ignore = compute_ptr_offset_of_batch;
+    ignore = block_2_ctile_map;
+
+#endif
+}
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t K1,
+          index_t M1PerThread,
+          index_t N1PerThread,
+          index_t KPerThread,
+          typename M1N1ThreadClusterM1Xs,
+          typename M1N1ThreadClusterN1Xs,
+          typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
+          typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+          typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
+          typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector,
+          enable_if_t<
+              is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+                  is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
+              bool> = false>
+struct DeviceBatchedGemmMultipleD_Dl : public DeviceBatchedGemmMultiD<ALayout,
+                                                                      BLayout,
+                                                                      DsLayout,
+                                                                      ELayout,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      DsDataType,
+                                                                      EDataType,
+                                                                      AElementwiseOperation,
+                                                                      BElementwiseOperation,
+                                                                      CDEElementwiseOperation>
+
+{
+    using DeviceOp                      = DeviceBatchedGemmMultipleD_Dl;
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr auto K1Number = Number<K1>{};
+
+    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
+    {
+        const index_t K0 = K / K1;
+
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(M, PadM)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
+    {
+        const index_t K0 = K / K1;
+
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    template <typename ELay>
+    static auto MakeEGridDescriptor_M_N(index_t M, index_t N, index_t StrideE)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideE));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                                         const std::array<index_t, NumDTensor>& NRaws,
+                                         const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return DeviceOp::MakeEGridDescriptor_M_N<DLayout>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
+    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
+    using DsGridDesc_M_N    = decltype(MakeDsGridDescriptor_M_N({}, {}, {}));
+    using EGridDesc_M_N     = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
+
+    struct ComputePtrOffsetOfStridedBatch
+    {
+        ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
+                                       index_t BatchStrideB,
+                                       std::array<ck::index_t, NumDTensor> BatchStrideDs,
+                                       index_t BatchStrideE)
+            : BatchStrideA_(BatchStrideA),
+              BatchStrideB_(BatchStrideB),
+              BatchStrideDs_(BatchStrideDs),
+              BatchStrideE_(BatchStrideE)
+        {
+        }
+
+        __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideA_);
+        }
+
+        __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideB_);
+        }
+
+        __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
+        {
+            std::array<long_index_t, NumDTensor> ds_offset;
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                ds_offset[i] = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]);
+            });
+            return ds_offset;
+        }
+
+        __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideE_);
+        }
+
+        private:
+        index_t BatchStrideA_;
+        index_t BatchStrideB_;
+        std::array<ck::index_t, NumDTensor> BatchStrideDs_;
+        index_t BatchStrideE_;
+    };
+
+    // GridwiseGemm
+    using GridwiseGemm =
+        GridwiseGemmDlMultipleD_km_kn_mn<BlockSize,
+                                         ADataType,
+                                         AccDataType,
+                                         DsDataType,
+                                         EDataType,
+                                         AElementwiseOperation,
+                                         BElementwiseOperation,
+                                         CDEElementwiseOperation,
+                                         InMemoryDataOperationEnum::Set,
+                                         AGridDesc_K0_M_K1,
+                                         BGridDesc_K0_N_K1,
+                                         EGridDesc_M_N,
+                                         MPerBlock,
+                                         NPerBlock,
+                                         K0PerBlock,
+                                         K1,
+                                         M1PerThread,
+                                         N1PerThread,
+                                         KPerThread,
+                                         M1N1ThreadClusterM1Xs,
+                                         M1N1ThreadClusterN1Xs,
+                                         ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterArrangeOrder,
+                                         ABlockTransferSrcAccessOrder,
+                                         ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+                                         ABlockTransferSrcVectorTensorContiguousDimOrder,
+                                         ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+                                         BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterArrangeOrder,
+                                         BBlockTransferSrcAccessOrder,
+                                         BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+                                         BBlockTransferSrcVectorTensorContiguousDimOrder,
+                                         BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+                                         CThreadTransferSrcDstAccessOrder,
+                                         CThreadTransferSrcDstVectorDim,
+                                         CThreadTransferDstScalarPerVector>;
+
+    using AGridDesc_K0_M0_M1_K1 =
+        decltype(GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
+    using BGridDesc_K0_N0_N1_K1 =
+        decltype(GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(BGridDesc_K0_N_K1{}));
+    using DsGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(DsGridDesc_M_N{}));
+    using EGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(EGridDesc_M_N{}));
+    using DefaultBlock2CTileMap =
+        decltype(GridwiseGemm::MakeDefaultBlock2CTileMap(EGridDesc_M_N{}));
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a_grid,
+                 const void* p_b_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 index_t M,
+                 index_t N,
+                 index_t K,
+                 index_t Batch,
+                 index_t StrideA,
+                 index_t StrideB,
+                 std::array<index_t, NumDTensor> StrideDs,
+                 index_t StrideE,
+                 index_t BatchStrideA,
+                 index_t BatchStrideB,
+                 const std::array<ck::index_t, NumDTensor>& BatchStrideDs,
+                 index_t BatchStrideE,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
+              p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              K_(K),
+              Batch_(Batch),
+              a_grid_desc_k0_m0_m1_k1_{},
+              b_grid_desc_k0_n0_n1_k1_{},
+              e_grid_desc_m0_m10_m11_n0_n10_n11_{},
+              compute_ptr_offset_of_batch_{BatchStrideA, BatchStrideB, BatchStrideDs, BatchStrideE},
+              block_2_ctile_map_{},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op}
+        {
+            a_grid_desc_k0_m_k1_ =
+                DeviceBatchedGemmMultipleD_Dl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
+            b_grid_desc_k0_n_k1_ =
+                DeviceBatchedGemmMultipleD_Dl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    DeviceOp::MakeEGridDescriptor_M_N<DLayout>(M, N, StrideDs[i]);
+            });
+            e_grid_desc_m_n_ =
+                DeviceBatchedGemmMultipleD_Dl::MakeEGridDescriptor_M_N<ELayout>(M, N, StrideE);
+
+            if(GridwiseGemm::CheckValidity(
+                   a_grid_desc_k0_m_k1_, b_grid_desc_k0_n_k1_, e_grid_desc_m_n_))
+            {
+                a_grid_desc_k0_m0_m1_k1_ =
+                    GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(a_grid_desc_k0_m_k1_);
+                b_grid_desc_k0_n0_n1_k1_ =
+                    GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(b_grid_desc_k0_n_k1_);
+
+                ds_grid_desc_m0_m10_m11_n0_n10_n11_ =
+                    GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(ds_grid_desc_m_n_);
+
+                e_grid_desc_m0_m10_m11_n0_n10_n11_ =
+                    GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(e_grid_desc_m_n_);
+
+                block_2_ctile_map_ = GridwiseGemm::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_);
+            }
+        }
+
+        //  private:
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseGemm::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        index_t K_;
+
+        // Batch
+        index_t Batch_;
+
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        AGridDesc_K0_M0_M1_K1 a_grid_desc_k0_m0_m1_k1_;
+        BGridDesc_K0_N0_N1_K1 b_grid_desc_k0_n0_n1_k1_;
+        DsGridDesc_M0_M10_M11_N0_N10_N11 ds_grid_desc_m0_m10_m11_n0_n10_n11_;
+        EGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11_;
+
+        // for calculating batch offset
+        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
+
+        DefaultBlock2CTileMap block_2_ctile_map_;
+
+        // TODO: unused since gridwise_gemm_dl_v1r3 does NOT support prologue for the time being.
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceBatchedGemmMultipleD_Dl::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            {
+                std::cout << "arg.a_grid_desc_k0_m0_m1_k1_{"
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I0) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.b_grid_desc_k0_n0_n1_k1_{"
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I0) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.e_grid_desc_m_n_{ " << arg.e_grid_desc_m_n_.GetLength(I0) << ", "
+                          << arg.e_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+            }
+
+            if(!GridwiseGemm::CheckValidity(
+                   arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemmDlMultipleD_km_kn_mn has invalid setting");
+            }
+
+            const index_t grid_size =
+                GridwiseGemm::CalculateGridSize(arg.e_grid_desc_m_n_.GetLength(I0),
+                                                arg.e_grid_desc_m_n_.GetLength(I1)) *
+                arg.Batch_;
+
+            auto launch_kernel = [&](auto has_main_k_block_loop,
+                                     auto has_double_tail_k_block_loop) {
+                constexpr bool has_main_loop   = has_main_k_block_loop.value;
+                constexpr bool has_double_loop = has_double_tail_k_block_loop.value;
+
+                const auto kernel =
+                    kernel_gemm_dl_multiple_d<GridwiseGemm,
+                                              ADataType,
+                                              typename GridwiseGemm::DsGridPointer,
+                                              EDataType,
+                                              AElementwiseOperation,
+                                              BElementwiseOperation,
+                                              CDEElementwiseOperation,
+                                              DeviceOp::AGridDesc_K0_M0_M1_K1,
+                                              DeviceOp::BGridDesc_K0_N0_N1_K1,
+                                              DeviceOp::DsGridDesc_M0_M10_M11_N0_N10_N11,
+                                              DeviceOp::EGridDesc_M0_M10_M11_N0_N10_N11,
+                                              ComputePtrOffsetOfStridedBatch,
+                                              DefaultBlock2CTileMap,
+                                              has_main_loop,
+                                              has_double_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_a_grid_,
+                                              arg.p_b_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              arg.Batch_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.a_grid_desc_k0_m0_m1_k1_,
+                                              arg.b_grid_desc_k0_n0_n1_k1_,
+                                              arg.ds_grid_desc_m0_m10_m11_n0_n10_n11_,
+                                              arg.e_grid_desc_m0_m10_m11_n0_n10_n11_,
+                                              arg.compute_ptr_offset_of_batch_,
+                                              arg.block_2_ctile_map_);
+            };
+
+            const auto K0                    = arg.a_grid_desc_k0_m0_m1_k1_.GetLength(I0);
+            const bool has_main_k_block_loop = GridwiseGemm::CalculateHasMainKBlockLoop(K0);
+            const bool has_double_tail_k_block_loop =
+                GridwiseGemm::CalculateHasDoubleTailKBlockLoop(K0);
+
+            if(has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, true>{});
+            }
+            else if(has_main_k_block_loop && !has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, false>{});
+            }
+            else if(!has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        // TODO: Enable for gfx90a after complier fix
+        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
+           ck::get_device_name() == "gfx1030" || ck::get_device_name() == "gfx940" ||
+           ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
+        {
+            bool pass = true;
+            pass      = pass && arg.K_ % K1 == 0;
+
+            pass = pass && GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                                       arg.b_grid_desc_k0_n_k1_,
+                                                       arg.e_grid_desc_m_n_);
+
+            return pass;
+        }
+        else
+        {
+            return false;
+        }
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_a,
+                             const void* p_b,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t Batch,
+                             index_t StrideA,
+                             index_t StrideB,
+                             std::array<ck::index_t, NumDTensor> StrideDs,
+                             index_t StrideE,
+                             index_t BatchStrideA,
+                             index_t BatchStrideB,
+                             const std::array<ck::index_t, NumDTensor>& BatchStrideDs,
+                             index_t BatchStrideE,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_e,
+                        M,
+                        N,
+                        K,
+                        Batch,
+                        StrideA,
+                        StrideB,
+                        StrideDs,
+                        StrideE,
+                        BatchStrideA,
+                        BatchStrideB,
+                        BatchStrideDs,
+                        BatchStrideE,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        const std::array<const void*, NumDTensor>& p_ds,
+                        void* p_e,
+                        index_t M,
+                        index_t N,
+                        index_t K,
+                        index_t Batch,
+                        index_t StrideA,
+                        index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor>& StrideDs,
+                        index_t StrideE,
+                        index_t BatchStrideA,
+                        index_t BatchStrideB,
+                        const std::array<ck::index_t, NumDTensor>& BatchStrideDs,
+                        index_t BatchStrideE,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          M,
+                                          N,
+                                          K,
+                                          Batch,
+                                          StrideA,
+                                          StrideB,
+                                          StrideDs,
+                                          StrideE,
+                                          BatchStrideA,
+                                          BatchStrideB,
+                                          BatchStrideDs,
+                                          BatchStrideE,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceBatchedGemmMultipleD_Dl"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << M1PerThread << ", "
+            << N1PerThread << ", "
+            << KPerThread
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
index 8142c9253b29b9dccc99287516a3d065119107b8..d259ebffae47f98f0e482b9d8b0e480d4a4079ed 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -68,7 +68,8 @@ __global__ void
             const index_t batch_count,
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -788,9 +789,45 @@ struct DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
         return true;
     }
 
+    // check if DsLayout is supported
+    template <typename RefLayout, typename DsLayout, const index_t NumDTensor>
+    static bool CheckDLayout()
+    {
+        static bool valid = true;
+        // iterate over DLayout tuple
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+            // if RefLayout and DLayout are same, keep valid true, otherwise false
+            valid = valid && is_same_v<RefLayout, DLayout>;
+        });
+        return valid;
+    }
+
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
+        {
+            return false;
+        }
+
+        // Check supported layouts
+        // A0 - Row
+        // B0 - Col
+        // D0s - Rows
+        // B1 - Row or Col
+        // D1s - Rows
+        // E1 - Row
+        if(!(is_same_v<tensor_layout::gemm::RowMajor, A0Layout> &&
+             is_same_v<tensor_layout::gemm::ColumnMajor, B0Layout> &&
+             CheckDLayout<tensor_layout::gemm::RowMajor, D0sLayout, NumD0Tensor>() &&
+             (is_same_v<tensor_layout::gemm::RowMajor, B1Layout> ||
+              is_same_v<tensor_layout::gemm::ColumnMajor,
+                        B1Layout>)&&CheckDLayout<tensor_layout::gemm::RowMajor,
+                                                 D1sLayout,
+                                                 NumD1Tensor>() &&
+             is_same_v<tensor_layout::gemm::RowMajor, E1Layout>))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
index 080e26ea89277c5d71362291381435fe1682b3b7..1d52416d4d8d875ac7e6dd9dcbf3ef8f98932b1c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -59,7 +59,8 @@ __global__ void
             const ComputeBasePrtOfBatch compute_base_ptr_of_batch_,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 59b6af1edbcdaf9ed1ffcdbc345fabacad76bede..73b12bfb4747f2b3971c9a30300c0e4f5745291e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -13,7 +13,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp"
 #include "ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
@@ -25,15 +25,17 @@ namespace device {
 template <typename GridwiseGemm,
           typename FloatAB,
           typename FloatC,
+          typename D0sPointer,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename AccElementwiseOperation,
+          typename C0DEElementwiseOperation,
           typename B1ElementwiseOperation,
-          typename CElementwiseOperation,
+          typename C1DEElementwiseOperation,
           typename AGridDesc_AK0_M_AK1,
           typename BGridDesc_BK0_N_BK1,
           typename B1GridDesc_BK0_N_BK1,
-          typename CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5,
           typename Block2CTileMap,
           typename ComputeBasePtrOfStridedBatch,
           typename C0MatrixMask,
@@ -47,22 +49,26 @@ __global__ void
             const FloatAB* __restrict__ p_b_grid,
             const FloatAB* __restrict__ p_b1_grid,
             FloatC* __restrict__ p_c_grid,
+            D0sPointer p_d0s_grid,
             const AElementwiseOperation a_element_op,
             const BElementwiseOperation b_element_op,
-            const AccElementwiseOperation acc_element_op,
+            const C0DEElementwiseOperation c0de_element_op,
             const B1ElementwiseOperation b1_element_op,
-            const CElementwiseOperation c_element_op,
+            const C1DEElementwiseOperation c1de_element_op,
             const AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1,
             const BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1,
             const B1GridDesc_BK0_N_BK1 b1_grid_desc_bk0_n_bk1,
-            const CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                c_grid_desc_mblock_mperblock_nblock_nperblock,
+            const C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                c1_grid_desc_mblock_mperblock_nblock_nperblock,
+            const D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5
+                d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
             const Block2CTileMap block_2_ctile_map,
             const index_t batch_count,
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch,
             const C0MatrixMask c0_matrix_mask)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -77,20 +83,28 @@ __global__ void
     const long_index_t c_batch_offset = __builtin_amdgcn_readfirstlane(
         static_cast<long_index_t>(compute_base_ptr_of_batch.GetCBasePtr(g_idx)));
 
+    static_for<0, p_d0s_grid.Size(), 1>{}([&](auto In) {
+        const long_index_t d0_batch_offset = __builtin_amdgcn_readfirstlane(
+            static_cast<long_index_t>(compute_base_ptr_of_batch.GetD0BasePtr(g_idx, In)));
+        p_d0s_grid(In) = p_d0s_grid(In) + d0_batch_offset;
+    });
+
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
                                                   p_b_grid + b_batch_offset,
                                                   p_b1_grid + b1_batch_offset,
                                                   p_c_grid + c_batch_offset,
+                                                  p_d0s_grid,
                                                   p_shared,
                                                   a_element_op,
                                                   b_element_op,
-                                                  acc_element_op,
+                                                  c0de_element_op,
                                                   b1_element_op,
-                                                  c_element_op,
+                                                  c1de_element_op,
                                                   a_grid_desc_ak0_m_ak1,
                                                   b_grid_desc_bk0_n_bk1,
                                                   b1_grid_desc_bk0_n_bk1,
-                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  c1_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
                                                   block_2_ctile_map,
                                                   c0_matrix_mask);
 #else
@@ -98,15 +112,17 @@ __global__ void
     ignore = p_b_grid;
     ignore = p_b1_grid;
     ignore = p_c_grid;
+    ignore = p_d0s_grid;
     ignore = a_element_op;
     ignore = b_element_op;
-    ignore = acc_element_op;
+    ignore = c0de_element_op;
     ignore = b1_element_op;
-    ignore = c_element_op;
+    ignore = c1de_element_op;
     ignore = a_grid_desc_ak0_m_ak1;
     ignore = b_grid_desc_bk0_n_bk1;
     ignore = b1_grid_desc_bk0_n_bk1;
-    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = c1_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5;
     ignore = block_2_ctile_map;
     ignore = batch_count;
     ignore = compute_base_ptr_of_batch;
@@ -126,15 +142,15 @@ template <index_t NumDimG,
           typename BDataType,
           typename B1DataType,
           typename CDataType,
-          typename Acc0BiasDataType,
-          typename Acc1BiasDataType,
+          typename D0sDataType,
+          typename D1sDataType,
           typename GemmAccDataType,
           typename CShuffleDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename AccElementwiseOperation,
+          typename C0DEElementwiseOperation,
           typename B1ElementwiseOperation,
-          typename CElementwiseOperation,
+          typename C1DEElementwiseOperation,
           GemmSpecialization GemmSpec,
           TensorSpecialization ASpec,
           TensorSpecialization BSpec,
@@ -181,7 +197,8 @@ template <index_t NumDimG,
           typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           MaskingSpecialization MaskingSpec,
-          LoopScheduler LoopSched = LoopScheduler::Default>
+          int D0sTransferSrcScalarPerVector = 4,
+          LoopScheduler LoopSched           = LoopScheduler::Default>
 struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
     : public DeviceBatchedGemmSoftmaxGemmPermute<NumDimG,
                                                  NumDimM,
@@ -192,23 +209,23 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                                                  BDataType,
                                                  B1DataType,
                                                  CDataType,
-                                                 Acc0BiasDataType,
-                                                 Acc1BiasDataType,
+                                                 D0sDataType,
+                                                 D1sDataType,
                                                  AElementwiseOperation,
                                                  BElementwiseOperation,
-                                                 AccElementwiseOperation,
+                                                 C0DEElementwiseOperation,
                                                  B1ElementwiseOperation,
-                                                 CElementwiseOperation,
+                                                 C1DEElementwiseOperation,
                                                  MaskingSpec>
 {
     static_assert(NumDimG > 0 && NumDimM > 0 && NumDimN > 0 && NumDimK > 0 && NumDimO > 0,
                   "Number of dimension must be greater than 0");
 
-    static constexpr index_t NumAcc0Bias = Acc0BiasDataType::Size();
-    static constexpr index_t NumAcc1Bias = Acc1BiasDataType::Size();
+    static constexpr index_t NumD0Tensor = D0sDataType::Size();
+    static constexpr index_t NumD1Tensor = D1sDataType::Size();
 
     // TODO ANT: implement bias combination
-    static_assert(NumAcc0Bias == 0 && NumAcc0Bias == 0, "Bias addition is unimplemented");
+    static_assert(NumD1Tensor == 0, "Gemm1 Bias addition is unimplemented");
 
 #if 0
     // TODO ANT: use alias
@@ -261,14 +278,40 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
             Number<B1K1>{});
     }
 
+    static auto MakeD0sGridDescriptor_M_N(
+        const std::array<std::vector<ck::index_t>, NumD0Tensor>& acc0_biases_gs_ms_ns_lengths,
+        const std::array<std::vector<ck::index_t>, NumD0Tensor>& acc0_biases_gs_ms_ns_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return Transform::MakeCGridDescriptor_M_N(acc0_biases_gs_ms_ns_lengths[i],
+                                                          acc0_biases_gs_ms_ns_strides[i]);
+            },
+            Number<NumD0Tensor>{});
+    }
+
+    static auto MakeD0sGridDescriptor_G_M_N(
+        const std::array<std::vector<ck::index_t>, NumD0Tensor>& acc0_biases_gs_ms_ns_lengths,
+        const std::array<std::vector<ck::index_t>, NumD0Tensor>& acc0_biases_gs_ms_ns_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return Transform::MakeCGridDescriptor_G_M_N(acc0_biases_gs_ms_ns_lengths[i],
+                                                            acc0_biases_gs_ms_ns_strides[i]);
+            },
+            Number<NumD0Tensor>{});
+    }
+
     using AGridDesc_AK0_M_AK1  = decltype(MakeAGridDescriptor_AK0_M_AK1({}, {}));
     using BGridDesc_BK0_N_BK1  = decltype(MakeBGridDescriptor_BK0_N_BK1({}, {}));
     using B1GridDesc_BK0_N_BK1 = decltype(MakeB1GridDescriptor_BK0_N_BK1({}, {}));
-    using CGridDesc_M_N        = decltype(Transform::MakeCGridDescriptor_M_N({}, {}));
+    using C1GridDesc_M_N       = decltype(Transform::MakeCGridDescriptor_M_N({}, {}));
     using AGridDesc_G_M_K      = decltype(Transform::MakeAGridDescriptor_G_M_K({}, {}));
     using BGridDesc_G_N_K      = decltype(Transform::MakeB0GridDescriptor_G_N_K({}, {}));
     using B1GridDesc_G_N_K     = decltype(Transform::MakeB1GridDescriptor_G_N_K({}, {}));
-    using CGridDesc_G_M_N      = decltype(Transform::MakeCGridDescriptor_G_M_N({}, {}));
+    using C1GridDesc_G_M_N     = decltype(Transform::MakeCGridDescriptor_G_M_N({}, {}));
+    using D0sGridDesc_M_N      = decltype(MakeD0sGridDescriptor_M_N({}, {}));
+    using D0sGridDesc_G_M_N    = decltype(MakeD0sGridDescriptor_G_M_N({}, {}));
 
     constexpr static auto make_MaskOutPredicate()
     {
@@ -288,11 +331,13 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         ComputeBasePtrOfStridedBatch(const AGridDesc_G_M_K& a_grid_desc_g_m_k,
                                      const BGridDesc_G_N_K& b_grid_desc_g_n_k,
                                      const B1GridDesc_G_N_K& b1_grid_desc_g_n_k,
-                                     const CGridDesc_G_M_N& c_grid_desc_g_m_n)
+                                     const C1GridDesc_G_M_N& c1_grid_desc_g_m_n,
+                                     const D0sGridDesc_G_M_N& d0s_grid_desc_g_m_n)
             : a_grid_desc_g_m_k_(a_grid_desc_g_m_k),
               b_grid_desc_g_n_k_(b_grid_desc_g_n_k),
               b1_grid_desc_g_n_k_(b1_grid_desc_g_n_k),
-              c_grid_desc_g_m_n_(c_grid_desc_g_m_n)
+              c1_grid_desc_g_m_n_(c1_grid_desc_g_m_n),
+              d0s_grid_desc_g_m_n_(d0s_grid_desc_g_m_n)
         {
         }
 
@@ -313,32 +358,42 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
 
         __host__ __device__ constexpr long_index_t GetCBasePtr(index_t g_idx) const
         {
-            return c_grid_desc_g_m_n_.CalculateOffset(make_multi_index(g_idx, 0, 0));
+            return c1_grid_desc_g_m_n_.CalculateOffset(make_multi_index(g_idx, 0, 0));
+        }
+
+        template <index_t I>
+        __host__ __device__ constexpr long_index_t GetD0BasePtr(index_t g_idx,
+                                                                Number<I> d0_idx) const
+        {
+            return d0s_grid_desc_g_m_n_[d0_idx].CalculateOffset(make_multi_index(g_idx, 0, 0));
         }
 
         private:
         AGridDesc_G_M_K a_grid_desc_g_m_k_;
         BGridDesc_G_N_K b_grid_desc_g_n_k_;
         B1GridDesc_G_N_K b1_grid_desc_g_n_k_;
-        CGridDesc_G_M_N c_grid_desc_g_m_n_;
+        C1GridDesc_G_M_N c1_grid_desc_g_m_n_;
+        D0sGridDesc_G_M_N d0s_grid_desc_g_m_n_;
     };
 
     // GridwiseGemm
-    using GridwiseGemm = GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle<
+    using GridwiseGemm = GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle<
         ADataType, // TODO: distinguish A/B datatype
         GemmAccDataType,
         CShuffleDataType,
         CDataType,
+        D0sDataType,
         AElementwiseOperation,
         BElementwiseOperation,
-        AccElementwiseOperation,
+        C0DEElementwiseOperation,
         B1ElementwiseOperation,
-        CElementwiseOperation,
+        C1DEElementwiseOperation,
         InMemoryDataOperationEnum::Set,
         AGridDesc_AK0_M_AK1,
         BGridDesc_BK0_N_BK1,
         B1GridDesc_BK0_N_BK1,
-        CGridDesc_M_N,
+        C1GridDesc_M_N,
+        D0sGridDesc_M_N,
         NumGemmKPrefetchStage,
         BlockSize,
         MPerBlock,
@@ -384,7 +439,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         CShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched,
         Transform::matrix_padder.PadN,
-        MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle>;
+        MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle,
+        D0sTransferSrcScalarPerVector>;
 
     // Argument
     // FIXME: constness
@@ -395,8 +451,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
             const BDataType* p_b_grid,
             const B1DataType* p_b1_grid,
             CDataType* p_c_grid,
-            const std::array<void*, NumAcc0Bias> p_acc0_biases,
-            const std::array<void*, NumAcc1Bias> p_acc1_biases,
+            const std::array<void*, NumD0Tensor> p_acc0_biases,
+            const std::array<void*, NumD1Tensor> p_acc1_biases,
             const std::vector<index_t>& a_gs_ms_ks_lengths,
             const std::vector<index_t>& a_gs_ms_ks_strides,
             const std::vector<index_t>& b_gs_ns_ks_lengths,
@@ -405,44 +461,48 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
             const std::vector<index_t>& b1_gs_gemm1ns_gemm1ks_strides, // b1_gs_os_ns_strides
             const std::vector<index_t>& c_gs_ms_gemm1ns_lengths,       // c_gs_ms_os_lengths
             const std::vector<index_t>& c_gs_ms_gemm1ns_strides,       // c_gs_ms_os_strides
-            const std::array<std::vector<ck::index_t>, NumAcc0Bias> acc0_biases_gs_ms_ns_lengths,
-            const std::array<std::vector<ck::index_t>, NumAcc0Bias> acc0_biases_gs_ms_ns_strides,
-            const std::array<std::vector<ck::index_t>, NumAcc1Bias>
+            const std::array<std::vector<ck::index_t>, NumD0Tensor>& acc0_biases_gs_ms_ns_lengths,
+            const std::array<std::vector<ck::index_t>, NumD0Tensor>& acc0_biases_gs_ms_ns_strides,
+            const std::array<std::vector<ck::index_t>, NumD1Tensor>&
                 acc1_biases_gs_ms_gemm1ns_lengths, // acc1_biases_gs_ms_os_lengths
-            const std::array<std::vector<ck::index_t>, NumAcc1Bias>
+            const std::array<std::vector<ck::index_t>, NumD1Tensor>&
                 acc1_biases_gs_ms_gemm1ns_strides, // acc1_biases_gs_ms_os_strides
             AElementwiseOperation a_element_op,
             BElementwiseOperation b_element_op,
-            AccElementwiseOperation acc_element_op,
+            C0DEElementwiseOperation c0de_element_op,
             B1ElementwiseOperation b1_element_op,
-            CElementwiseOperation c_element_op)
+            C1DEElementwiseOperation c1de_element_op)
             : p_a_grid_{p_a_grid},
               p_b_grid_{p_b_grid},
               p_b1_grid_{p_b1_grid},
               p_c_grid_{p_c_grid},
+              p_d0s_grid_{},
               a_grid_desc_ak0_m_ak1_{
                   DeviceOp::MakeAGridDescriptor_AK0_M_AK1(a_gs_ms_ks_lengths, a_gs_ms_ks_strides)},
               b_grid_desc_bk0_n_bk1_{
                   DeviceOp::MakeBGridDescriptor_BK0_N_BK1(b_gs_ns_ks_lengths, b_gs_ns_ks_strides)},
               b1_grid_desc_bk0_n_bk1_{DeviceOp::MakeB1GridDescriptor_BK0_N_BK1(
                   b1_gs_gemm1ns_gemm1ks_lengths, b1_gs_gemm1ns_gemm1ks_strides)},
-              c_grid_desc_m_n_{Transform::MakeCGridDescriptor_M_N(c_gs_ms_gemm1ns_lengths,
-                                                                  c_gs_ms_gemm1ns_strides)},
+              c1_grid_desc_m_n_{Transform::MakeCGridDescriptor_M_N(c_gs_ms_gemm1ns_lengths,
+                                                                   c_gs_ms_gemm1ns_strides)},
               a_grid_desc_g_m_k_{
                   Transform::MakeAGridDescriptor_G_M_K(a_gs_ms_ks_lengths, a_gs_ms_ks_strides)},
               b_grid_desc_g_n_k_{
                   Transform::MakeB0GridDescriptor_G_N_K(b_gs_ns_ks_lengths, b_gs_ns_ks_strides)},
               b1_grid_desc_g_n_k_{Transform::MakeB1GridDescriptor_G_N_K(
                   b1_gs_gemm1ns_gemm1ks_lengths, b1_gs_gemm1ns_gemm1ks_strides)},
-              c_grid_desc_g_m_n_{Transform::MakeCGridDescriptor_G_M_N(c_gs_ms_gemm1ns_lengths,
-                                                                      c_gs_ms_gemm1ns_strides)},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_ctile_map_{GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_)},
+              c1_grid_desc_g_m_n_{Transform::MakeCGridDescriptor_G_M_N(c_gs_ms_gemm1ns_lengths,
+                                                                       c_gs_ms_gemm1ns_strides)},
+              d0s_grid_desc_g_m_n_{DeviceOp::MakeD0sGridDescriptor_G_M_N(
+                  acc0_biases_gs_ms_ns_lengths, acc0_biases_gs_ms_ns_strides)},
+              c1_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              d0s_grid_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5_{},
+              block_2_ctile_map_{GridwiseGemm::MakeDefaultBlock2CTileMap(c1_grid_desc_m_n_)},
               a_element_op_{a_element_op},
               b_element_op_{b_element_op},
-              acc_element_op_{acc_element_op},
+              c0de_element_op_{c0de_element_op},
               b1_element_op_{b1_element_op},
-              c_element_op_{c_element_op},
+              c1de_element_op_{c1de_element_op},
               c0_matrix_mask_{b_grid_desc_g_n_k_.GetLength(I1)},
               raw_lengths_mz_nz_kz_gemm1nz_{a_gs_ms_ks_lengths[NumDimG + NumDimM - 1],
                                             b_gs_ns_ks_lengths[NumDimG + NumDimN - 1],
@@ -456,27 +516,44 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                                 b1_gs_gemm1ns_gemm1ks_strides[NumDimG + NumDimO + NumDimN - 1]},
               c_mz_gemm1nz_strides_{c_gs_ms_gemm1ns_strides[NumDimG + NumDimM - 1],
                                     c_gs_ms_gemm1ns_strides[NumDimG + NumDimM + NumDimO - 1]},
-              batch_count_{c_grid_desc_g_m_n_.GetLength(I0)},
-              compute_base_ptr_of_batch_{
-                  a_grid_desc_g_m_k_, b_grid_desc_g_n_k_, b1_grid_desc_g_n_k_, c_grid_desc_g_m_n_}
+              batch_count_{c1_grid_desc_g_m_n_.GetLength(I0)},
+              compute_base_ptr_of_batch_{a_grid_desc_g_m_k_,
+                                         b_grid_desc_g_n_k_,
+                                         b1_grid_desc_g_n_k_,
+                                         c1_grid_desc_g_m_n_,
+                                         d0s_grid_desc_g_m_n_}
         {
             // TODO ANT: implement bias addition
-            ignore = p_acc0_biases;
             ignore = p_acc1_biases;
-            ignore = acc0_biases_gs_ms_ns_lengths;
-            ignore = acc0_biases_gs_ms_ns_strides;
             ignore = acc1_biases_gs_ms_gemm1ns_lengths;
             ignore = acc1_biases_gs_ms_gemm1ns_strides;
 
+            static_for<0, NumD0Tensor, 1>{}([&](auto i) {
+                using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
+                // D0 pointer
+                p_d0s_grid_(i) = static_cast<const D0DataType*>(p_acc0_biases[i]);
+                // for  check
+                d0s_nl_ns_lengths_strides_[i].push_back(
+                    acc0_biases_gs_ms_ns_lengths[i][NumDimG + NumDimM]);
+                d0s_nl_ns_lengths_strides_[i].push_back(
+                    acc0_biases_gs_ms_ns_strides[i][NumDimG + NumDimM]);
+            });
+
             if(GridwiseGemm::CheckValidity(a_grid_desc_ak0_m_ak1_,
                                            b_grid_desc_bk0_n_bk1_,
                                            b1_grid_desc_bk0_n_bk1_,
-                                           c_grid_desc_m_n_,
+                                           c1_grid_desc_m_n_,
                                            block_2_ctile_map_))
             {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        c_grid_desc_m_n_);
+                c1_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeC1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        c1_grid_desc_m_n_);
+
+                D0sGridDesc_M_N d0s_grid_desc_m_n{DeviceOp::MakeD0sGridDescriptor_M_N(
+                    acc0_biases_gs_ms_ns_lengths, acc0_biases_gs_ms_ns_strides)};
+                d0s_grid_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5_ =
+                    GridwiseGemm::MakeD0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(
+                        d0s_grid_desc_m_n);
             }
         }
 
@@ -491,9 +568,9 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
             std::cout << "b1_grid_desc_g_n_k_: " << b1_grid_desc_g_n_k_.GetLength(I0) << ", "
                       << b1_grid_desc_g_n_k_.GetLength(I1) << ", "
                       << b1_grid_desc_g_n_k_.GetLength(I2) << '\n';
-            std::cout << "c_grid_desc_g_m_n_: " << c_grid_desc_g_m_n_.GetLength(I0) << ", "
-                      << c_grid_desc_g_m_n_.GetLength(I1) << ", "
-                      << c_grid_desc_g_m_n_.GetLength(I2) << '\n';
+            std::cout << "c1_grid_desc_g_m_n_: " << c1_grid_desc_g_m_n_.GetLength(I0) << ", "
+                      << c1_grid_desc_g_m_n_.GetLength(I1) << ", "
+                      << c1_grid_desc_g_m_n_.GetLength(I2) << '\n';
         }
 
         // pointers
@@ -501,18 +578,23 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         const BDataType* p_b_grid_;
         const B1DataType* p_b1_grid_;
         CDataType* p_c_grid_;
+        typename GridwiseGemm::D0sGridPointer p_d0s_grid_;
 
         // tensor descriptor
         AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
         BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
         B1GridDesc_BK0_N_BK1 b1_grid_desc_bk0_n_bk1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
+        C1GridDesc_M_N c1_grid_desc_m_n_;
         AGridDesc_G_M_K a_grid_desc_g_m_k_;
         BGridDesc_G_N_K b_grid_desc_g_n_k_;
         B1GridDesc_G_N_K b1_grid_desc_g_n_k_;
-        CGridDesc_G_M_N c_grid_desc_g_m_n_;
-        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            c_grid_desc_mblock_mperblock_nblock_nperblock_;
+        C1GridDesc_G_M_N c1_grid_desc_g_m_n_;
+        D0sGridDesc_G_M_N d0s_grid_desc_g_m_n_;
+
+        typename GridwiseGemm::C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            c1_grid_desc_mblock_mperblock_nblock_nperblock_;
+        typename GridwiseGemm::D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5
+            d0s_grid_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5_;
 
         // block-to-c-tile map
         typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
@@ -520,9 +602,9 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         // element-wise op
         AElementwiseOperation a_element_op_;
         BElementwiseOperation b_element_op_;
-        AccElementwiseOperation acc_element_op_;
+        C0DEElementwiseOperation c0de_element_op_;
         B1ElementwiseOperation b1_element_op_;
-        CElementwiseOperation c_element_op_;
+        C1DEElementwiseOperation c1de_element_op_;
 
         // check C0 masking and padding
         C0MatrixMask c0_matrix_mask_;
@@ -533,6 +615,7 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         std::vector<index_t> b_nz_kz_strides_;
         std::vector<index_t> b1_nz_kz_strides_;
         std::vector<index_t> c_mz_gemm1nz_strides_;
+        std::array<std::vector<ck::index_t>, NumD0Tensor> d0s_nl_ns_lengths_strides_;
 
         index_t batch_count_;
         ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch_;
@@ -551,7 +634,7 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
             }
 
             const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_) * arg.batch_count_;
+                arg.block_2_ctile_map_.CalculateGridSize(arg.c1_grid_desc_m_n_) * arg.batch_count_;
 
             // Gemm0_K
             const auto K =
@@ -564,15 +647,17 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                     GridwiseGemm,
                     ADataType, // TODO: distiguish A/B datatype
                     CDataType,
+                    typename GridwiseGemm::D0sGridPointer,
                     AElementwiseOperation,
                     BElementwiseOperation,
-                    AccElementwiseOperation,
+                    C0DEElementwiseOperation,
                     B1ElementwiseOperation,
-                    CElementwiseOperation,
+                    C1DEElementwiseOperation,
                     DeviceOp::AGridDesc_AK0_M_AK1,
                     DeviceOp::BGridDesc_BK0_N_BK1,
                     DeviceOp::B1GridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                    typename GridwiseGemm::C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                    typename GridwiseGemm::D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5,
                     typename GridwiseGemm::DefaultBlock2CTileMap,
                     ComputeBasePtrOfStridedBatch,
                     C0MatrixMask,
@@ -587,15 +672,17 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                                               arg.p_b_grid_,
                                               arg.p_b1_grid_,
                                               arg.p_c_grid_,
+                                              arg.p_d0s_grid_,
                                               arg.a_element_op_,
                                               arg.b_element_op_,
-                                              arg.acc_element_op_,
+                                              arg.c0de_element_op_,
                                               arg.b1_element_op_,
-                                              arg.c_element_op_,
+                                              arg.c1de_element_op_,
                                               arg.a_grid_desc_ak0_m_ak1_,
                                               arg.b_grid_desc_bk0_n_bk1_,
                                               arg.b1_grid_desc_bk0_n_bk1_,
-                                              arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.c1_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.d0s_grid_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5_,
                                               arg.block_2_ctile_map_,
                                               arg.batch_count_,
                                               arg.compute_base_ptr_of_batch_,
@@ -636,7 +723,9 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         arg.Print();
 #endif
 
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
@@ -644,9 +733,9 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         // TODO ANT: Check if tensor specialization & strides mismatch
 
         // Check if C permute dimension matches GEMM + GEMM shape
-        const index_t c_g       = arg.c_grid_desc_g_m_n_.GetLength(I0); // unpadded
-        const index_t c_m       = arg.c_grid_desc_m_n_.GetLength(I0);
-        const index_t c_gemm1n  = arg.c_grid_desc_m_n_.GetLength(I1);
+        const index_t c_g       = arg.c1_grid_desc_g_m_n_.GetLength(I0); // unpadded
+        const index_t c_m       = arg.c1_grid_desc_m_n_.GetLength(I0);
+        const index_t c_gemm1n  = arg.c1_grid_desc_m_n_.GetLength(I1);
         const index_t a_m       = arg.a_grid_desc_ak0_m_ak1_.GetLength(I1);
         const index_t b1_gemm1n = arg.b1_grid_desc_bk0_n_bk1_.GetLength(I1);
 
@@ -692,11 +781,23 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         {
             return false;
         }
+        for(int i = 0; i < NumD0Tensor; i++)
+        {
+            if(arg.d0s_nl_ns_lengths_strides_[i][1] == 1 &&
+               arg.d0s_nl_ns_lengths_strides_[i][0] % D0sTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+            if(arg.d0s_nl_ns_lengths_strides_[i][1] != 1 && D0sTransferSrcScalarPerVector != 1)
+            {
+                return false;
+            }
+        }
 
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
                                            arg.b_grid_desc_bk0_n_bk1_,
                                            arg.b1_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_m_n_,
+                                           arg.c1_grid_desc_m_n_,
                                            arg.block_2_ctile_map_);
     }
 
@@ -711,8 +812,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         const BDataType* p_b,
         const B1DataType* p_b1,
         CDataType* p_c,
-        const std::array<void*, NumAcc0Bias> p_acc0_biases,
-        const std::array<void*, NumAcc1Bias> p_acc1_biases,
+        const std::array<void*, NumD0Tensor> p_acc0_biases,
+        const std::array<void*, NumD1Tensor> p_acc1_biases,
         const std::vector<index_t>& a_gs_ms_ks_lengths,
         const std::vector<index_t>& a_gs_ms_ks_strides,
         const std::vector<index_t>& b_gs_ns_ks_lengths,
@@ -721,17 +822,17 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         const std::vector<index_t>& b1_gs_gemm1ns_gemm1ks_strides, // b1_gs_os_ns_strides
         const std::vector<index_t>& c_gs_ms_gemm1ns_lengths,       // c_gs_ms_os_lengths
         const std::vector<index_t>& c_gs_ms_gemm1ns_strides,       // c_gs_ms_os_strides
-        const std::array<std::vector<ck::index_t>, NumAcc0Bias> acc0_biases_gs_ms_ns_lengths,
-        const std::array<std::vector<ck::index_t>, NumAcc0Bias> acc0_biases_gs_ms_ns_strides,
-        const std::array<std::vector<ck::index_t>, NumAcc1Bias>
+        const std::array<std::vector<ck::index_t>, NumD0Tensor> acc0_biases_gs_ms_ns_lengths,
+        const std::array<std::vector<ck::index_t>, NumD0Tensor> acc0_biases_gs_ms_ns_strides,
+        const std::array<std::vector<ck::index_t>, NumD1Tensor>
             acc1_biases_gs_ms_gemm1ns_lengths, // acc1_biases_gs_ms_os_lengths
-        const std::array<std::vector<ck::index_t>, NumAcc1Bias>
+        const std::array<std::vector<ck::index_t>, NumD1Tensor>
             acc1_biases_gs_ms_gemm1ns_strides, // acc1_biases_gs_ms_os_strides
         AElementwiseOperation a_element_op,
         BElementwiseOperation b_element_op,
-        AccElementwiseOperation acc_element_op,
+        C0DEElementwiseOperation c0de_element_op,
         B1ElementwiseOperation b1_element_op,
-        CElementwiseOperation c_element_op)
+        C1DEElementwiseOperation c1de_element_op)
     {
         return Argument{p_a,
                         p_b,
@@ -753,9 +854,9 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                         acc1_biases_gs_ms_gemm1ns_strides, // acc1_biases_gs_ms_os_strides
                         a_element_op,
                         b_element_op,
-                        acc_element_op,
+                        c0de_element_op,
                         b1_element_op,
-                        c_element_op};
+                        c1de_element_op};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -767,8 +868,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         const void* p_b,
         const void* p_b1,
         void* p_c,
-        const std::array<void*, NumAcc0Bias> p_acc0_biases,
-        const std::array<void*, NumAcc1Bias> p_acc1_biases,
+        const std::array<void*, NumD0Tensor> p_acc0_biases,
+        const std::array<void*, NumD1Tensor> p_acc1_biases,
         const std::vector<index_t>& a_gs_ms_ks_lengths,
         const std::vector<index_t>& a_gs_ms_ks_strides,
         const std::vector<index_t>& b_gs_ns_ks_lengths,
@@ -777,17 +878,17 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         const std::vector<index_t>& b1_gs_gemm1ns_gemm1ks_strides, // b1_gs_os_ns_strides
         const std::vector<index_t>& c_gs_ms_gemm1ns_lengths,       // c_gs_ms_os_lengths
         const std::vector<index_t>& c_gs_ms_gemm1ns_strides,       // c_gs_ms_os_strides
-        const std::array<std::vector<ck::index_t>, NumAcc0Bias> acc0_biases_gs_ms_ns_lengths,
-        const std::array<std::vector<ck::index_t>, NumAcc0Bias> acc0_biases_gs_ms_ns_strides,
-        const std::array<std::vector<ck::index_t>, NumAcc1Bias>
+        const std::array<std::vector<ck::index_t>, NumD0Tensor> acc0_biases_gs_ms_ns_lengths,
+        const std::array<std::vector<ck::index_t>, NumD0Tensor> acc0_biases_gs_ms_ns_strides,
+        const std::array<std::vector<ck::index_t>, NumD1Tensor>
             acc1_biases_gs_ms_gemm1ns_lengths, // acc1_biases_gs_ms_os_lengths
-        const std::array<std::vector<ck::index_t>, NumAcc1Bias>
+        const std::array<std::vector<ck::index_t>, NumD1Tensor>
             acc1_biases_gs_ms_gemm1ns_strides, // acc1_biases_gs_ms_os_strides
         AElementwiseOperation a_element_op,
         BElementwiseOperation b_element_op,
-        AccElementwiseOperation acc_element_op,
+        C0DEElementwiseOperation c0de_element_op,
         B1ElementwiseOperation b1_element_op,
-        CElementwiseOperation c_element_op) override
+        C1DEElementwiseOperation c1de_element_op) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
@@ -809,9 +910,9 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                                           acc1_biases_gs_ms_gemm1ns_strides,
                                           a_element_op,
                                           b_element_op,
-                                          acc_element_op,
+                                          c0de_element_op,
                                           b1_element_op,
-                                          c_element_op);
+                                          c1de_element_op);
     }
 
     // polymorphic
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
index 1f21f2d7122ca7b31553c915f6698f249c3f0aaf..2405640ca94bf0aae26d99b327e4d228313cafd9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -62,7 +62,8 @@ __global__ void
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch,
             const C0MatrixMask c0_matrix_mask)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -612,7 +613,9 @@ struct DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
index 48a224456052fd98184601db84428eb2271fa695..b6a0567fee5277bab30a74a9160591144bae60ae 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -45,74 +45,46 @@ namespace device {
  * realize BatchedGemm and GroupedGemm (and the corresponding GEMM fusion).
  *
  */
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_K0_M_K1,
-          typename BGridDesc_K0_N_K1,
-          typename CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename ComputePtrOffsetOfBatch,
-          typename Block2CTileMap,
-          bool HasMainKBlockLoop>
+template <typename DeviceOp, typename GridwiseGemm, bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_batched_gemm_xdlops_v2r3(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const index_t batch_count,
-            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
-            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
-            const CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-            const AElementwiseOperation a_element_op,
-            const BElementwiseOperation b_element_op,
-            const CElementwiseOperation c_element_op,
-            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
-            const Block2CTileMap block_2_ctile_map)
+        kernel_batched_gemm_xdlops_v2r3(const typename DeviceOp::Argument karg)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
-        __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+        __builtin_amdgcn_readfirstlane(get_grid_size() / karg.Batch);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
 
     const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
-        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
+        static_cast<long_index_t>(karg.compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
     const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
-        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
+        static_cast<long_index_t>(karg.compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
     const long_index_t c_batch_offset = __builtin_amdgcn_readfirstlane(
-        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetCPtrOffset(g_idx)));
+        static_cast<long_index_t>(karg.compute_ptr_offset_of_batch.GetCPtrOffset(g_idx)));
 
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
-                                                  p_b_grid + b_batch_offset,
-                                                  p_c_grid + c_batch_offset,
+    const auto a_grid_desc_k0_m_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeAGridDescriptor_K0_M_K1(
+            karg.M, karg.MPadded, karg.K, karg.K0, karg.StrideA));
+    const auto b_grid_desc_k0_n_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeBGridDescriptor_K0_N_K1(
+            karg.K, karg.N, karg.NPadded, karg.K0, karg.StrideB));
+    const auto c_grid_desc_m_n = amd_wave_read_first_lane(GridwiseGemm::MakeCGridDescriptor_M_N(
+        karg.M, karg.MPadded, karg.N, karg.NPadded, karg.StrideC));
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(karg.p_a_grid + a_batch_offset,
+                                                  karg.p_b_grid + b_batch_offset,
+                                                  karg.p_c_grid + c_batch_offset,
                                                   p_shared,
                                                   a_grid_desc_k0_m_k1,
                                                   b_grid_desc_k0_n_k1,
-                                                  c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  block_2_ctile_map);
+                                                  c_grid_desc_m_n);
 #else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_c_grid;
-    ignore = batch_count;
-    ignore = a_grid_desc_k0_m_k1;
-    ignore = b_grid_desc_k0_n_k1;
-    ignore = c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = compute_ptr_offset_of_batch;
-    ignore = block_2_ctile_map;
+    ignore = karg;
 #endif
 }
 
@@ -170,93 +142,6 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
 
     static constexpr auto K1Number = Number<K1>{};
 
-    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
-    {
-        assert(K % K1 == 0);
-
-        const index_t K0 = K / K1;
-
-        const auto a_grid_desc_m_k = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-
-        const auto a_grid_desc_k0_mp_k1 =
-            transform_tensor_descriptor(a_grid_desc_m_k,
-                                        make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                                                   make_right_pad_transform(M, PadM)),
-                                        make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-        return a_grid_desc_k0_mp_k1;
-    }
-
-    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
-    {
-        assert(K % K1 == 0);
-
-        const index_t K0 = K / K1;
-
-        const auto b_grid_desc_k_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
-            }
-        }();
-
-        const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-        const auto b_grid_desc_k0_np_k1 =
-            transform_tensor_descriptor(b_grid_desc_k_n,
-                                        make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                                                   make_right_pad_transform(N, PadN)),
-                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-        return b_grid_desc_k0_np_k1;
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
-    {
-        const auto c_grid_desc_m_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
-            }
-        }();
-
-        const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-        const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-        const auto c_grid_desc_mp_np = transform_tensor_descriptor(
-            c_grid_desc_m_n,
-            make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        return c_grid_desc_mp_np;
-    }
-
-    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
-    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
-    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
     struct ComputePtrOffsetOfStridedBatch
     {
         ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
@@ -288,121 +173,82 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
     };
 
     // GridwiseGemm
-    using GridwiseGemm =
-        GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<BlockSize,
-                                                ADataType, // TODO: distinguish A/B datatype
-                                                AccDataType,
-                                                CDataType,
-                                                InMemoryDataOperationEnum::Set,
-                                                AGridDesc_K0_M_K1,
-                                                BGridDesc_K0_N_K1,
-                                                CGridDesc_M_N,
-                                                AElementwiseOperation,
-                                                BElementwiseOperation,
-                                                CElementwiseOperation,
-                                                MPerBlock,
-                                                NPerBlock,
-                                                K0PerBlock,
-                                                MPerXDL,
-                                                NPerXDL,
-                                                K1,
-                                                MXdlPerWave,
-                                                NXdlPerWave,
-                                                ABlockTransferThreadClusterLengths_K0_M_K1,
-                                                ABlockTransferThreadClusterArrangeOrder,
-                                                ABlockTransferSrcAccessOrder,
-                                                ABlockTransferSrcVectorDim,
-                                                ABlockTransferSrcScalarPerVector,
-                                                ABlockTransferDstScalarPerVector_K1,
-                                                false, // AThreadTransferSrcResetCoordinateAfterRun,
-                                                ABlockLdsAddExtraM,
-                                                BBlockTransferThreadClusterLengths_K0_N_K1,
-                                                BBlockTransferThreadClusterArrangeOrder,
-                                                BBlockTransferSrcAccessOrder,
-                                                BBlockTransferSrcVectorDim,
-                                                BBlockTransferSrcScalarPerVector,
-                                                BBlockTransferDstScalarPerVector_K1,
-                                                false, // BThreadTransferSrcResetCoordinateAfterRun,
-                                                BBlockLdsAddExtraN,
-                                                Sequence<2, 3, 0, 1, 7, 5, 4, 6>,
-                                                CThreadTransferSrcDstVectorDim,
-                                                CThreadTransferDstScalarPerVector,
-                                                NumGemmKPrefetchStage,
-                                                LoopSched,
-                                                PipelineVer>;
-
-    using CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 =
-        decltype(GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}));
-    using Block2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
+    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext<
+        BlockSize,
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        CDataType,
+        InMemoryDataOperationEnum::Set,
+        ALayout,
+        BLayout,
+        CLayout,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        GemmSpecialization::MNPadding,
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerXDL,
+        NPerXDL,
+        K1,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        Sequence<2, 3, 0, 1, 7, 5, 4, 6>,
+        CThreadTransferSrcDstVectorDim,
+        CThreadTransferDstScalarPerVector,
+        NumGemmKPrefetchStage,
+        LoopSched,
+        PipelineVer>;
+
+    using Problem = typename GridwiseGemm::Problem;
 
     // Argument
-    struct Argument : public BaseArgument
+    struct Argument : public Problem, public BaseArgument
     {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t M,
-                 index_t N,
-                 index_t K,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
+        Argument(const ADataType* p_a_grid_,
+                 const BDataType* p_b_grid_,
+                 CDataType* p_c_grid_,
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_,
                  index_t BatchStrideA,
                  index_t BatchStrideB,
                  index_t BatchStrideC,
-                 index_t Batch,
-                 index_t M01,
-                 index_t N01,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              Batch_(Batch),
-              a_grid_desc_k0_m_k1_{
-                  DeviceBatchedGemmXdl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA)},
-              b_grid_desc_k0_n_k1_{
-                  DeviceBatchedGemmXdl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB)},
-              c_grid_desc_m_n_{DeviceBatchedGemmXdl::MakeCGridDescriptor_M_N(M, N, StrideC)},
-              c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
-              compute_ptr_offset_of_batch_{BatchStrideA, BatchStrideB, BatchStrideC},
-              block_2_ctile_map_{
-                  GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_, M01, N01)},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              kraw_{K}
+                 index_t Batch_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid{p_a_grid_},
+              p_b_grid{p_b_grid_},
+              p_c_grid{p_c_grid_},
+              Batch(Batch_),
+              compute_ptr_offset_of_batch{BatchStrideA, BatchStrideB, BatchStrideC}
         {
-            if(GridwiseGemm::CheckValidity(a_grid_desc_k0_m_k1_,
-                                           b_grid_desc_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
-                    GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
-            }
         }
 
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        index_t Batch_;
-        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
-        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
-        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
-        Block2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t kraw_;
+        const ADataType* p_a_grid;
+        const BDataType* p_b_grid;
+        CDataType* p_c_grid;
+        index_t Batch;
+        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch;
     };
 
     // Invoker
@@ -410,107 +256,39 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
     {
         using Argument = DeviceBatchedGemmXdl::Argument;
 
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
         {
-#if DEBUG_LOG
+            if(stream_config.log_level_ > 0)
             {
-                std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
-                          << ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
-                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
-                          << ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
-                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.c_grid_desc_m_n_{" << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                          << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+                karg.Print();
             }
-#endif
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                            arg.b_grid_desc_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(karg))
             {
                 throw std::runtime_error(
-                    "wrong! GridwiseBatchedGemm_km_kn_m0m1n0n1_xdlops_v2r3 has invalid setting");
+                    "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_) * arg.Batch_;
-
-            const auto K =
-                arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
+            auto [gdx, gdy, gdz] = GridwiseGemm::CalculateGridSize(karg.M, karg.N);
+            gdx *= karg.Batch;
 
             float ave_time = 0;
 
-            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(karg.K))
             {
-                const auto kernel = kernel_batched_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    ComputePtrOffsetOfStridedBatch,
-                    remove_reference_t<Block2CTileMap>,
-                    true>;
-
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.Batch_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.compute_ptr_offset_of_batch_,
-                                                  arg.block_2_ctile_map_);
+                const auto kernel =
+                    kernel_batched_gemm_xdlops_v2r3<DeviceBatchedGemmXdl, GridwiseGemm, true>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
             else
             {
-                const auto kernel = kernel_batched_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    ComputePtrOffsetOfStridedBatch,
-                    remove_reference_t<Block2CTileMap>,
-                    false>;
-
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.Batch_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.compute_ptr_offset_of_batch_,
-                                                  arg.block_2_ctile_map_);
+                const auto kernel =
+                    kernel_batched_gemm_xdlops_v2r3<DeviceBatchedGemmXdl, GridwiseGemm, false>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
 
             return ave_time;
@@ -530,17 +308,14 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
         return true;
     }
 
-    static bool IsSupportedArgument(const Argument& arg)
+    static bool IsSupportedArgument(const Problem& problem)
     {
-        if(arg.kraw_ % K1 != 0)
+        if(problem.K % K1 != 0)
         {
             return false;
         }
 
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                           arg.b_grid_desc_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(problem);
     }
 
     // polymorphic
@@ -561,10 +336,7 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                              index_t BatchStrideA,
                              index_t BatchStrideB,
                              index_t BatchStrideC,
-                             index_t Batch,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             index_t Batch)
     {
         return Argument{p_a,
                         p_b,
@@ -578,12 +350,7 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                         BatchStrideA,
                         BatchStrideB,
                         BatchStrideC,
-                        Batch,
-                        1,
-                        1,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+                        Batch};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -602,9 +369,9 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                                                       index_t BatchStrideB,
                                                       index_t BatchStrideC,
                                                       index_t Batch,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op) override
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
@@ -618,12 +385,7 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                                           BatchStrideA,
                                           BatchStrideB,
                                           BatchStrideC,
-                                          Batch,
-                                          1,
-                                          1,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          Batch);
     }
 
     // polymorphic
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp
index ab16a757f699bb4e4ba781873bbf87538633d9c1..f46237e005642d433c29f5ab0f027a224043dbef 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
index 5a16ff765bbc5c0bc1b68d80262c88a0c8f84416..ad8e7956033db31db2992d6fa9f86cf16a29b150 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -10,12 +10,14 @@
 #include "ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
 #include "ck/tensor_operation/gpu/device/welford_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp"
 #include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp"
-#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/hip_check_error.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -114,8 +116,8 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
 
     static auto MakeMeanVarCountOutputMG2dDescriptor(int invariantLength, int blkGroupSize)
     {
-        const auto grid_desc_m_g =
-            make_naive_tensor_descriptor_packed(make_tuple(invariantLength, blkGroupSize));
+        const auto grid_desc_m_g = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, blkGroupSize), make_tuple(1, invariantLength));
 
         const auto mPad =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
@@ -132,9 +134,9 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
 
     static auto MakeMeanVarCountInputMK2dDescriptor(int invariantLength, int blkGroupSize)
     {
-        const auto reduceLength = blkGroupSize;
-        const auto grid_desc_m_k =
-            make_naive_tensor_descriptor_packed(make_tuple(invariantLength, reduceLength));
+        const auto reduceLength  = blkGroupSize;
+        const auto grid_desc_m_k = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, reduceLength), make_tuple(1, invariantLength));
 
         const auto mPad =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
@@ -244,8 +246,8 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
                     int testBlkGroupSize = (reduce_length_ + (K_BlockTileSize * iterations) - 1) /
                                            (K_BlockTileSize * iterations);
 
-                    // we want the blkGroupSize be not more than 128
-                    if(testBlkGroupSize <= 128)
+                    // we want the blkGroupSize be not more than 16
+                    if(testBlkGroupSize <= 16)
                         break;
 
                     iterations++;
@@ -319,6 +321,8 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
         void* workspace_mean_;
         void* workspace_variance_;
         void* workspace_count_;
+
+        void* control_;
     };
 
     size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
@@ -340,6 +344,11 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
             // workspace for welford intermediate count
             workspace_size +=
                 pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(int32_t) + 64;
+
+            // workspace for barrier objects, each barrier object consists of two integers
+            // TODO: allocate barrier object memory globally to reuse it by other operators
+            workspace_size += (pArg_->invariant_length_ + M_BlockTileSize - 1) / M_BlockTileSize *
+                              sizeof(int) * 2;
         }
 
         return (workspace_size);
@@ -353,7 +362,6 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
 
         if(UseMultiblockInK && pArg_->blkGroupSize_ > 1)
         {
-
             // setup buffer used for intermediate welford mean
             pArg_->workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
 
@@ -374,6 +382,18 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
             // setup buffer used for intermediate welfor count
             pArg_->workspace_count_ =
                 reinterpret_cast<char*>(pArg_->workspace_variance_) + variance_space_sz;
+
+            index_t count_space_sz =
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(int32_t);
+
+            count_space_sz = math::integer_least_multiple(count_space_sz, 64);
+
+            pArg_->control_ = reinterpret_cast<char*>(pArg_->workspace_count_) + count_space_sz;
+
+            index_t control_space_sz = (pArg_->invariant_length_ + M_BlockTileSize - 1) /
+                                       M_BlockTileSize * sizeof(int) * 2;
+
+            hip_check_error(hipMemset(pArg_->control_, 0, control_space_sz));
         };
     };
 
@@ -402,6 +422,32 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
                 using MeanVarCountGridDesc_M_G = decltype(mean_var_count_grid_desc_m_g);
                 using MeanVarCountGridDesc_M_K = decltype(mean_var_count_grid_desc_m_k);
 
+                using GridwiseMultiblockBatchNormForward_ =
+                    GridwiseMultiblockBatchNormForward<XDataType,
+                                                       YDataType,
+                                                       AccDataType,
+                                                       ScaleDataType,
+                                                       BiasDataType,
+                                                       MeanVarDataType,
+                                                       YElementwiseOp,
+                                                       XYGridDesc_M_K,
+                                                       MeanVarCountGridDesc_M_G,
+                                                       MeanVarCountGridDesc_M_K,
+                                                       ScaleBiasMeanVarGridDesc_M,
+                                                       ScaleBiasMeanVarGridDesc_M,
+                                                       GetReduceCountPerThreadFunctor,
+                                                       BlockSize,
+                                                       MThreadClusterSize,
+                                                       KThreadClusterSize,
+                                                       MThreadSliceSize,
+                                                       KThreadSliceSize,
+                                                       XSrcYDstVectorDim,
+                                                       XSrcVectorSize,
+                                                       YDstVectorSize,
+                                                       ScaleSrcVectorSize,
+                                                       BiasSrcVectorSize,
+                                                       MeanVarSrcDstVectorSize>;
+
                 using GridwiseMultiblockWelfordFirstHalf_ =
                     GridwiseMultiblockWelfordFirstHalf<XDataType,
                                                        AccDataType,
@@ -441,78 +487,136 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
                                                                    BiasSrcVectorSize,
                                                                    MeanVarSrcDstVectorSize>;
 
-                index_t numMeanVarCountBlockTileIteration =
-                    (arg.blkGroupSize_ + KThreadClusterSize - 1) / KThreadClusterSize;
-
-                const auto kern_multiblock_welford_first_half =
-                    kernel_multiblock_welford_first_half<GridwiseMultiblockWelfordFirstHalf_,
-                                                         XDataType,
-                                                         MeanVarDataType,
-                                                         XYGridDesc_M_K,
-                                                         MeanVarCountGridDesc_M_G,
-                                                         GetReduceCountPerThreadFunctor>;
-
-                const auto kern_welford_second_half_batchnorm_forward_final =
-                    kernel_welford_second_half_batchnorm_forward_final<
-                        GridwiseWelfordSecondHalfBatchNormForwardFinal_,
-                        XDataType,
-                        YDataType,
-                        AccDataType,
-                        ScaleDataType,
-                        BiasDataType,
-                        MeanVarDataType,
-                        YElementwiseOp,
-                        XYGridDesc_M_K,
-                        MeanVarCountGridDesc_M_K,
-                        ScaleBiasMeanVarGridDesc_M,
-                        ScaleBiasMeanVarGridDesc_M>;
-
-                avg_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kern_multiblock_welford_first_half,
-                                           dim3(arg.gridSize_),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.x_grid_desc_m_k_,
-                                           mean_var_count_grid_desc_m_g,
-                                           get_reduce_count_per_thread,
-                                           arg.numBlockTileIteration_,
-                                           arg.p_x_,
-                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
-                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
-                                           static_cast<int32_t*>(arg.workspace_count_));
-
-                avg_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kern_welford_second_half_batchnorm_forward_final,
-                                           dim3(arg.gridSize_),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.x_grid_desc_m_k_,
-                                           arg.y_grid_desc_m_k_,
-                                           mean_var_count_grid_desc_m_k,
-                                           arg.scale_grid_desc_m_,
-                                           arg.bias_grid_desc_m_,
-                                           arg.mean_var_grid_desc_m_,
-                                           arg.blkGroupSize_,
-                                           arg.numBlockTileIteration_,
-                                           numMeanVarCountBlockTileIteration,
-                                           arg.epsilon_,
-                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
-                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
-                                           static_cast<int32_t*>(arg.workspace_count_),
-                                           arg.p_x_,
-                                           arg.p_scale_,
-                                           arg.p_bias_,
-                                           arg.y_elementwise_op_,
-                                           arg.p_y_,
-                                           arg.updateMovingAverage_,
-                                           arg.averageFactor_,
-                                           arg.resultRunningMean_,
-                                           arg.resultRunningVariance_,
-                                           arg.saveMeanInvVariance_,
-                                           arg.resultSaveMean_,
-                                           arg.resultSaveInvVariance_);
+                // It is found that:
+                // 1) gfx1030 does not support the GLC enabled vector load/store, so using the
+                //    two-kernel method for gfx1030
+                // 2) Profiler on gfx908 could hang even though it works when running examples
+                // 3) Single-kernel method works on gfx1100, but the performance it not better
+                //    than two-kernel method (due to more warps participating the barrier)
+                if(ck::get_device_name() == "gfx90a")
+                {
+                    const auto kern_multiblock_batchnorm_fwd_ =
+                        kernel_multiblock_batchnorm_forward<GridwiseMultiblockBatchNormForward_,
+                                                            XDataType,
+                                                            YDataType,
+                                                            AccDataType,
+                                                            ScaleDataType,
+                                                            BiasDataType,
+                                                            MeanVarDataType,
+                                                            YElementwiseOp,
+                                                            XYGridDesc_M_K,
+                                                            MeanVarCountGridDesc_M_G,
+                                                            MeanVarCountGridDesc_M_K,
+                                                            ScaleBiasMeanVarGridDesc_M,
+                                                            ScaleBiasMeanVarGridDesc_M,
+                                                            GetReduceCountPerThreadFunctor>;
+
+                    avg_time += launch_and_time_kernel(
+                        stream_config,
+                        kern_multiblock_batchnorm_fwd_,
+                        dim3(arg.gridSize_),
+                        dim3(BlockSize),
+                        0,
+                        arg.x_grid_desc_m_k_,
+                        arg.y_grid_desc_m_k_,
+                        mean_var_count_grid_desc_m_g, // for writing to mean/variance/count
+                                                      // workspace by multiple workgroups
+                        mean_var_count_grid_desc_m_k, // for reading from mean/variance/count
+                                                      // workspace by each workgroup
+                        arg.scale_grid_desc_m_,
+                        arg.bias_grid_desc_m_,
+                        arg.mean_var_grid_desc_m_,
+                        get_reduce_count_per_thread,
+                        arg.numBlockTileIteration_,
+                        arg.epsilon_,
+                        arg.p_x_,
+                        static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                        static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                        static_cast<int32_t*>(arg.workspace_count_),
+                        static_cast<int*>(arg.control_),
+                        arg.p_scale_,
+                        arg.p_bias_,
+                        arg.y_elementwise_op_,
+                        arg.p_y_,
+                        arg.updateMovingAverage_, // true or false
+                        arg.averageFactor_,
+                        arg.resultRunningMean_,
+                        arg.resultRunningVariance_,
+                        arg.saveMeanInvVariance_, // true or false
+                        arg.resultSaveMean_,
+                        arg.resultSaveInvVariance_);
+                }
+                else
+                {
+                    const auto kern_multiblock_welford_first_half =
+                        kernel_multiblock_welford_first_half<GridwiseMultiblockWelfordFirstHalf_,
+                                                             XDataType,
+                                                             MeanVarDataType,
+                                                             XYGridDesc_M_K,
+                                                             MeanVarCountGridDesc_M_G,
+                                                             GetReduceCountPerThreadFunctor>;
+
+                    const auto kern_welford_second_half_batchnorm_forward_final =
+                        kernel_welford_second_half_batchnorm_forward_final<
+                            GridwiseWelfordSecondHalfBatchNormForwardFinal_,
+                            XDataType,
+                            YDataType,
+                            AccDataType,
+                            ScaleDataType,
+                            BiasDataType,
+                            MeanVarDataType,
+                            YElementwiseOp,
+                            XYGridDesc_M_K,
+                            MeanVarCountGridDesc_M_K,
+                            ScaleBiasMeanVarGridDesc_M,
+                            ScaleBiasMeanVarGridDesc_M>;
+
+                    avg_time += launch_and_time_kernel(
+                        stream_config,
+                        kern_multiblock_welford_first_half,
+                        dim3(arg.gridSize_),
+                        dim3(BlockSize),
+                        0,
+                        arg.x_grid_desc_m_k_,
+                        mean_var_count_grid_desc_m_g,
+                        get_reduce_count_per_thread,
+                        arg.numBlockTileIteration_,
+                        arg.p_x_,
+                        static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                        static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                        static_cast<int32_t*>(arg.workspace_count_));
+
+                    avg_time += launch_and_time_kernel(
+                        stream_config,
+                        kern_welford_second_half_batchnorm_forward_final,
+                        dim3(arg.gridSize_),
+                        dim3(BlockSize),
+                        0,
+                        arg.x_grid_desc_m_k_,
+                        arg.y_grid_desc_m_k_,
+                        mean_var_count_grid_desc_m_k,
+                        arg.scale_grid_desc_m_,
+                        arg.bias_grid_desc_m_,
+                        arg.mean_var_grid_desc_m_,
+                        arg.blkGroupSize_,
+                        arg.numBlockTileIteration_,
+                        arg.epsilon_,
+                        static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                        static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                        static_cast<int32_t*>(arg.workspace_count_),
+                        arg.p_x_,
+                        arg.p_scale_,
+                        arg.p_bias_,
+                        arg.y_elementwise_op_,
+                        arg.p_y_,
+                        arg.updateMovingAverage_,
+                        arg.averageFactor_,
+                        arg.resultRunningMean_,
+                        arg.resultRunningVariance_,
+                        arg.saveMeanInvVariance_,
+                        arg.resultSaveMean_,
+                        arg.resultSaveInvVariance_);
+                };
             }
             else
             {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..b826793c27be946a99d0d96a99d745fd6c618822
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp
@@ -0,0 +1,714 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/reduction_operator.hpp"
+#include "ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
+#include "ck/tensor_operation/gpu/device/welford_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename XDataType,
+          typename YDataType,
+          typename AccDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          typename YElementwiseOp,
+          index_t Rank,
+          index_t NumBatchNormReduceDim,
+          bool UseMultiblockInK,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcYDstVectorDim,
+          index_t XSrcVectorSize,
+          index_t YDstVectorSize,
+          index_t ScaleSrcVectorSize,
+          index_t BiasSrcVectorSize,
+          index_t MeanVarSrcDstVectorSize>
+struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
+                                                          YDataType,
+                                                          AccDataType,
+                                                          ScaleDataType,
+                                                          BiasDataType,
+                                                          MeanVarDataType,
+                                                          YElementwiseOp,
+                                                          Rank,
+                                                          NumBatchNormReduceDim>
+{
+    static_assert(Rank <= 6, "Bigger Rank size is not supported!");
+    static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize,
+                  "Invalid thread cluster size assignments!");
+
+    static_assert((XSrcYDstVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcYDstVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static constexpr index_t NumInvariantDim = Rank - NumBatchNormReduceDim;
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static auto MakeXY2dDescriptor(const std::array<index_t, Rank>& xyLengths,
+                                   const std::array<index_t, Rank>& xyStrides,
+                                   int blkGroupSize,
+                                   int numBlockTileIteration)
+    {
+        const auto tupleXYLengths =
+            generate_tuple([&](auto I) { return xyLengths[I]; }, Number<Rank>{});
+        const auto tupleXYStrides =
+            generate_tuple([&](auto I) { return xyStrides[I]; }, Number<Rank>{});
+
+        const auto raw_grid_desc = make_naive_tensor_descriptor(tupleXYLengths, tupleXYStrides);
+
+        const auto grid_desc_m_k = [&]() {
+            using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
+            using ReduceDims    = typename arithmetic_sequence_gen<NumInvariantDim, Rank, 1>::type;
+
+            const auto reduceDimLengths =
+                generate_tuple([&](auto I) { return xyLengths[NumInvariantDim + I]; },
+                               Number<NumBatchNormReduceDim>{});
+            const auto invariantDimLengths =
+                generate_tuple([&](auto I) { return xyLengths[I]; }, Number<NumInvariantDim>{});
+
+            return transform_tensor_descriptor(raw_grid_desc,
+                                               make_tuple(make_merge_transform(invariantDimLengths),
+                                                          make_merge_transform(reduceDimLengths)),
+                                               make_tuple(InvariantDims{}, ReduceDims{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }();
+
+        const auto invariantLength = grid_desc_m_k.GetLength(Number<0>{});
+        const auto reduceLength    = grid_desc_m_k.GetLength(Number<1>{});
+
+        const int workSizePerBlock = K_BlockTileSize * numBlockTileIteration;
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+        const auto kPad = workSizePerBlock * blkGroupSize - reduceLength;
+
+        auto grid_desc_m_k_padded =
+            transform_tensor_descriptor(grid_desc_m_k,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad),
+                                                   make_right_pad_transform(reduceLength, kPad)),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (grid_desc_m_k_padded);
+    };
+
+    static auto MakeMeanVarCountOutputMG2dDescriptor(int invariantLength, int blkGroupSize)
+    {
+        const auto grid_desc_m_g = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, blkGroupSize), make_tuple(1, invariantLength));
+
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+
+        auto grid_desc_m_g_padded =
+            transform_tensor_descriptor(grid_desc_m_g,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad),
+                                                   make_pass_through_transform(blkGroupSize)),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (grid_desc_m_g_padded);
+    };
+
+    static auto MakeMeanVarCountInputMK2dDescriptor(int invariantLength, int blkGroupSize)
+    {
+        const auto reduceLength  = blkGroupSize;
+        const auto grid_desc_m_k = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, reduceLength), make_tuple(1, invariantLength));
+
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+        const auto kPad =
+            math::integer_least_multiple(reduceLength, KThreadClusterSize) - reduceLength;
+
+        auto grid_desc_m_k_padded =
+            transform_tensor_descriptor(grid_desc_m_k,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad),
+                                                   make_right_pad_transform(reduceLength, kPad)),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (grid_desc_m_k_padded);
+    };
+
+    static auto
+    MakeScaleBiasMeanVar1dDescriptor(const std::array<index_t, NumInvariantDim>& lengths,
+                                     const std::array<index_t, NumInvariantDim>& strides)
+    {
+        const auto tupleLengths =
+            generate_tuple([&](auto I) { return lengths[I]; }, Number<NumInvariantDim>{});
+        const auto tupleStrides =
+            generate_tuple([&](auto I) { return strides[I]; }, Number<NumInvariantDim>{});
+
+        auto raw_grid_desc = make_naive_tensor_descriptor(tupleLengths, tupleStrides);
+
+        auto grid_desc_m = transform_tensor_descriptor(
+            raw_grid_desc,
+            make_tuple(make_merge_transform(tupleLengths)),
+            make_tuple(typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type{}),
+            make_tuple(Sequence<0>{}));
+
+        const auto invariantLength = grid_desc_m.GetLength(Number<0>{});
+
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+
+        auto grid_desc_m_padded =
+            transform_tensor_descriptor(grid_desc_m,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+        return (grid_desc_m_padded);
+    };
+
+    using XYGridDesc_M_K             = decltype(MakeXY2dDescriptor({1}, {1}, 1, 1));
+    using ScaleBiasMeanVarGridDesc_M = decltype(MakeScaleBiasMeanVar1dDescriptor({1}, {1}));
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const std::array<index_t, Rank> xyLengths,
+                 const std::array<index_t, Rank> xStrides,
+                 const std::array<index_t, Rank> yStrides,
+                 const std::array<int, NumBatchNormReduceDim> reduceDims,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleBiasMeanVarLengths,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleStrides,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnBiasStrides,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnMeanVarStrides,
+                 const XDataType* p_x,
+                 const ScaleDataType* p_scale,
+                 const BiasDataType* p_bias,
+                 const YElementwiseOp y_elementwise_op,
+                 double epsilon,
+                 YDataType* p_y,
+                 MeanVarDataType* resultSaveMean,
+                 MeanVarDataType* resultSaveInvVariance,
+                 double averageFactor,
+                 MeanVarDataType* resultRunningMean,
+                 MeanVarDataType* resultRunningVariance)
+            : bnScaleBiasMeanVarLengths_(bnScaleBiasMeanVarLengths),
+              bnScaleStrides_(bnScaleStrides),
+              bnBiasStrides_(bnBiasStrides),
+              bnMeanVarStrides_(bnMeanVarStrides),
+              p_x_(p_x),
+              p_scale_(p_scale),
+              p_bias_(p_bias),
+              y_elementwise_op_(y_elementwise_op),
+              p_y_(p_y),
+              resultSaveMean_(resultSaveMean),
+              resultSaveInvVariance_(resultSaveInvVariance),
+              resultRunningMean_(resultRunningMean),
+              resultRunningVariance_(resultRunningVariance)
+        {
+            xyLengths_ =
+                shuffle_tensor_dimensions<Rank, NumBatchNormReduceDim>(xyLengths, reduceDims);
+            xStrides_ =
+                shuffle_tensor_dimensions<Rank, NumBatchNormReduceDim>(xStrides, reduceDims);
+            yStrides_ =
+                shuffle_tensor_dimensions<Rank, NumBatchNormReduceDim>(yStrides, reduceDims);
+
+            std::tie(invariant_length_, reduce_length_) =
+                get_2d_lengths<Rank, NumBatchNormReduceDim>(xyLengths_);
+
+            epsilon_       = type_convert<AccDataType>(epsilon);
+            averageFactor_ = type_convert<AccDataType>(averageFactor);
+
+            updateMovingAverage_ =
+                (resultRunningMean != nullptr && resultRunningVariance != nullptr);
+            saveMeanInvVariance_ = (resultSaveMean != nullptr && resultSaveInvVariance_ != nullptr);
+
+            if(UseMultiblockInK)
+            {
+                int iterations = 1;
+                while(true)
+                {
+                    int testBlkGroupSize = (reduce_length_ + (K_BlockTileSize * iterations) - 1) /
+                                           (K_BlockTileSize * iterations);
+
+                    // we want the blkGroupSize be not more than 16
+                    if(testBlkGroupSize <= 16)
+                        break;
+
+                    iterations++;
+                };
+
+                blkGroupSize_ = (reduce_length_ + (K_BlockTileSize * iterations) - 1) /
+                                (K_BlockTileSize * iterations);
+
+                numBlockTileIteration_ = iterations;
+            }
+            else
+            {
+                blkGroupSize_          = 1;
+                numBlockTileIteration_ = (reduce_length_ + K_BlockTileSize - 1) / K_BlockTileSize;
+            };
+
+            gridSize_ = (invariant_length_ + M_BlockTileSize - 1) / M_BlockTileSize * blkGroupSize_;
+
+            x_grid_desc_m_k_ =
+                MakeXY2dDescriptor(xyLengths_, xStrides_, blkGroupSize_, numBlockTileIteration_);
+            y_grid_desc_m_k_ =
+                MakeXY2dDescriptor(xyLengths_, yStrides_, blkGroupSize_, numBlockTileIteration_);
+            scale_grid_desc_m_ =
+                MakeScaleBiasMeanVar1dDescriptor(bnScaleBiasMeanVarLengths, bnScaleStrides_);
+            bias_grid_desc_m_ =
+                MakeScaleBiasMeanVar1dDescriptor(bnScaleBiasMeanVarLengths, bnBiasStrides_);
+            mean_var_grid_desc_m_ =
+                MakeScaleBiasMeanVar1dDescriptor(bnScaleBiasMeanVarLengths, bnMeanVarStrides_);
+        }
+
+        AccDataType epsilon_;
+        AccDataType averageFactor_;
+
+        bool updateMovingAverage_;
+        bool saveMeanInvVariance_;
+
+        std::array<index_t, Rank> xyLengths_;
+        std::array<index_t, Rank> xStrides_;
+        std::array<index_t, Rank> yStrides_;
+
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleBiasMeanVarLengths_;
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleStrides_;
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnBiasStrides_;
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnMeanVarStrides_;
+
+        const XDataType* p_x_;
+        const ScaleDataType* p_scale_;
+        const BiasDataType* p_bias_;
+        const YElementwiseOp y_elementwise_op_;
+        YDataType* p_y_;
+
+        MeanVarDataType* resultSaveMean_;
+        MeanVarDataType* resultSaveInvVariance_;
+
+        MeanVarDataType* resultRunningMean_;
+        MeanVarDataType* resultRunningVariance_;
+
+        long_index_t invariant_length_;
+        long_index_t reduce_length_;
+
+        int blkGroupSize_;
+        int numBlockTileIteration_;
+        size_t gridSize_;
+
+        XYGridDesc_M_K x_grid_desc_m_k_;
+        XYGridDesc_M_K y_grid_desc_m_k_;
+        ScaleBiasMeanVarGridDesc_M scale_grid_desc_m_;
+        ScaleBiasMeanVarGridDesc_M bias_grid_desc_m_;
+        ScaleBiasMeanVarGridDesc_M mean_var_grid_desc_m_;
+
+        void* workspace_mean_;
+        void* workspace_variance_;
+        void* workspace_count_;
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        size_t workspace_size = 0;
+
+        if(UseMultiblockInK && pArg_->blkGroupSize_ > 1)
+        {
+            // workspace for welford intermediate mean
+            workspace_size +=
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType) + 64;
+
+            // workspace for welford intermediate variance
+            workspace_size +=
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType) + 64;
+
+            // workspace for welford intermediate count
+            workspace_size +=
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(int32_t) + 64;
+        }
+
+        return (workspace_size);
+    };
+
+    void SetWorkSpacePointer(BaseArgument* pArg, void* p_workspace) const override
+    {
+        Argument* pArg_ = dynamic_cast<Argument*>(pArg);
+
+        pArg_->p_workspace_ = p_workspace;
+
+        if(UseMultiblockInK && pArg_->blkGroupSize_ > 1)
+        {
+
+            // setup buffer used for intermediate welford mean
+            pArg_->workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
+
+            index_t mean_space_sz =
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType);
+
+            mean_space_sz = math::integer_least_multiple(mean_space_sz, 64);
+
+            // setup buffer used for intermediate welford varirance
+            pArg_->workspace_variance_ =
+                reinterpret_cast<char*>(pArg_->workspace_mean_) + mean_space_sz;
+
+            index_t variance_space_sz =
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType);
+
+            variance_space_sz = math::integer_least_multiple(variance_space_sz, 64);
+
+            // setup buffer used for intermediate welfor count
+            pArg_->workspace_count_ =
+                reinterpret_cast<char*>(pArg_->workspace_variance_) + variance_space_sz;
+        };
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            float avg_time = 0;
+
+            if(UseMultiblockInK && arg.blkGroupSize_ > 1)
+            {
+                using GetReduceCountPerThreadFunctor =
+                    GetReduceCountPerThreadForMultiblockWelford<K_BlockTileSize, KThreadSliceSize>;
+
+                GetReduceCountPerThreadFunctor get_reduce_count_per_thread(
+                    arg.blkGroupSize_, arg.numBlockTileIteration_, arg.reduce_length_);
+
+                const auto mean_var_count_grid_desc_m_g =
+                    DeviceBatchNormFwdImpl::MakeMeanVarCountOutputMG2dDescriptor(
+                        arg.invariant_length_, arg.blkGroupSize_);
+
+                const auto mean_var_count_grid_desc_m_k =
+                    DeviceBatchNormFwdImpl::MakeMeanVarCountInputMK2dDescriptor(
+                        arg.invariant_length_, arg.blkGroupSize_);
+
+                using MeanVarCountGridDesc_M_G = decltype(mean_var_count_grid_desc_m_g);
+                using MeanVarCountGridDesc_M_K = decltype(mean_var_count_grid_desc_m_k);
+
+                using GridwiseMultiblockWelfordFirstHalf_ =
+                    GridwiseMultiblockWelfordFirstHalf<XDataType,
+                                                       AccDataType,
+                                                       MeanVarDataType,
+                                                       XYGridDesc_M_K,
+                                                       MeanVarCountGridDesc_M_G,
+                                                       GetReduceCountPerThreadFunctor,
+                                                       BlockSize,
+                                                       MThreadClusterSize,
+                                                       KThreadClusterSize,
+                                                       MThreadSliceSize,
+                                                       KThreadSliceSize,
+                                                       XSrcYDstVectorDim,
+                                                       XSrcVectorSize>;
+
+                using GridwiseWelfordSecondHalfBatchNormForwardFinal_ =
+                    GridwiseWelfordSecondHalfBatchNormForwardFinal<XDataType,
+                                                                   YDataType,
+                                                                   AccDataType,
+                                                                   ScaleDataType,
+                                                                   BiasDataType,
+                                                                   MeanVarDataType,
+                                                                   YElementwiseOp,
+                                                                   XYGridDesc_M_K,
+                                                                   MeanVarCountGridDesc_M_K,
+                                                                   ScaleBiasMeanVarGridDesc_M,
+                                                                   ScaleBiasMeanVarGridDesc_M,
+                                                                   BlockSize,
+                                                                   MThreadClusterSize,
+                                                                   KThreadClusterSize,
+                                                                   MThreadSliceSize,
+                                                                   KThreadSliceSize,
+                                                                   XSrcYDstVectorDim,
+                                                                   XSrcVectorSize,
+                                                                   YDstVectorSize,
+                                                                   ScaleSrcVectorSize,
+                                                                   BiasSrcVectorSize,
+                                                                   MeanVarSrcDstVectorSize>;
+
+                const auto kern_multiblock_welford_first_half =
+                    kernel_multiblock_welford_first_half<GridwiseMultiblockWelfordFirstHalf_,
+                                                         XDataType,
+                                                         MeanVarDataType,
+                                                         XYGridDesc_M_K,
+                                                         MeanVarCountGridDesc_M_G,
+                                                         GetReduceCountPerThreadFunctor>;
+
+                const auto kern_welford_second_half_batchnorm_forward_final =
+                    kernel_welford_second_half_batchnorm_forward_final<
+                        GridwiseWelfordSecondHalfBatchNormForwardFinal_,
+                        XDataType,
+                        YDataType,
+                        AccDataType,
+                        ScaleDataType,
+                        BiasDataType,
+                        MeanVarDataType,
+                        YElementwiseOp,
+                        XYGridDesc_M_K,
+                        MeanVarCountGridDesc_M_K,
+                        ScaleBiasMeanVarGridDesc_M,
+                        ScaleBiasMeanVarGridDesc_M>;
+
+                avg_time +=
+                    launch_and_time_kernel(stream_config,
+                                           kern_multiblock_welford_first_half,
+                                           dim3(arg.gridSize_),
+                                           dim3(BlockSize),
+                                           0,
+                                           arg.x_grid_desc_m_k_,
+                                           mean_var_count_grid_desc_m_g,
+                                           get_reduce_count_per_thread,
+                                           arg.numBlockTileIteration_,
+                                           arg.p_x_,
+                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                                           static_cast<int32_t*>(arg.workspace_count_));
+
+                avg_time +=
+                    launch_and_time_kernel(stream_config,
+                                           kern_welford_second_half_batchnorm_forward_final,
+                                           dim3(arg.gridSize_),
+                                           dim3(BlockSize),
+                                           0,
+                                           arg.x_grid_desc_m_k_,
+                                           arg.y_grid_desc_m_k_,
+                                           mean_var_count_grid_desc_m_k,
+                                           arg.scale_grid_desc_m_,
+                                           arg.bias_grid_desc_m_,
+                                           arg.mean_var_grid_desc_m_,
+                                           arg.blkGroupSize_,
+                                           arg.numBlockTileIteration_,
+                                           arg.epsilon_,
+                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                                           static_cast<int32_t*>(arg.workspace_count_),
+                                           arg.p_x_,
+                                           arg.p_scale_,
+                                           arg.p_bias_,
+                                           arg.y_elementwise_op_,
+                                           arg.p_y_,
+                                           arg.updateMovingAverage_,
+                                           arg.averageFactor_,
+                                           arg.resultRunningMean_,
+                                           arg.resultRunningVariance_,
+                                           arg.saveMeanInvVariance_,
+                                           arg.resultSaveMean_,
+                                           arg.resultSaveInvVariance_);
+            }
+            else
+            {
+                using GetReduceCountPerThreadFunctor =
+                    GetReduceCountPerThreadForBlockwiseWelford<K_BlockTileSize, KThreadSliceSize>;
+
+                GetReduceCountPerThreadFunctor get_reduce_count_per_thread(
+                    arg.numBlockTileIteration_, arg.reduce_length_);
+
+                using GridwiseBatchNormForwardWithBlockwiseWelford_ =
+                    GridwiseBatchNormForwardWithBlockwiseWelford<XDataType,
+                                                                 YDataType,
+                                                                 AccDataType,
+                                                                 ScaleDataType,
+                                                                 BiasDataType,
+                                                                 MeanVarDataType,
+                                                                 YElementwiseOp,
+                                                                 XYGridDesc_M_K,
+                                                                 ScaleBiasMeanVarGridDesc_M,
+                                                                 ScaleBiasMeanVarGridDesc_M,
+                                                                 GetReduceCountPerThreadFunctor,
+                                                                 BlockSize,
+                                                                 MThreadClusterSize,
+                                                                 KThreadClusterSize,
+                                                                 MThreadSliceSize,
+                                                                 KThreadSliceSize,
+                                                                 XSrcYDstVectorDim,
+                                                                 XSrcVectorSize,
+                                                                 YDstVectorSize,
+                                                                 ScaleSrcVectorSize,
+                                                                 BiasSrcVectorSize,
+                                                                 MeanVarSrcDstVectorSize>;
+
+                const auto kern_batchnorm_fwd = kernel_batchnorm_forward_with_blockwise_welford<
+                    GridwiseBatchNormForwardWithBlockwiseWelford_,
+                    XDataType,
+                    YDataType,
+                    AccDataType,
+                    ScaleDataType,
+                    BiasDataType,
+                    MeanVarDataType,
+                    YElementwiseOp,
+                    XYGridDesc_M_K,
+                    ScaleBiasMeanVarGridDesc_M,
+                    ScaleBiasMeanVarGridDesc_M,
+                    GetReduceCountPerThreadFunctor>;
+
+                avg_time += launch_and_time_kernel(stream_config,
+                                                   kern_batchnorm_fwd,
+                                                   dim3(arg.gridSize_),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.x_grid_desc_m_k_,
+                                                   arg.y_grid_desc_m_k_,
+                                                   arg.scale_grid_desc_m_,
+                                                   arg.bias_grid_desc_m_,
+                                                   arg.mean_var_grid_desc_m_,
+                                                   get_reduce_count_per_thread,
+                                                   arg.numBlockTileIteration_,
+                                                   arg.epsilon_,
+                                                   arg.p_x_,
+                                                   arg.p_scale_,
+                                                   arg.p_bias_,
+                                                   arg.y_elementwise_op_,
+                                                   arg.p_y_,
+                                                   arg.updateMovingAverage_, // true or false
+                                                   arg.averageFactor_,
+                                                   arg.resultRunningMean_,
+                                                   arg.resultRunningVariance_,
+                                                   arg.saveMeanInvVariance_, // true or false
+                                                   arg.resultSaveMean_,
+                                                   arg.resultSaveInvVariance_);
+            };
+
+            return (avg_time);
+        };
+
+        float Run(const BaseArgument* pArg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(pArg), stream_config);
+        };
+    };
+
+    bool IsSupportedArgument(const BaseArgument* pArg) override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        if constexpr(XSrcYDstVectorDim == 0)
+        {
+            if(pArg_->xStrides_[NumInvariantDim - 1] != 1 ||
+               pArg_->yStrides_[NumInvariantDim - 1] != 1)
+                return false;
+
+            if(pArg_->xyLengths_[NumInvariantDim - 1] % XSrcVectorSize != 0 ||
+               pArg_->xyLengths_[NumInvariantDim - 1] % YDstVectorSize != 0)
+                return false;
+        }
+        else
+        {
+            if(pArg_->xStrides_[Rank - 1] != 1 || pArg_->yStrides_[Rank - 1] != 1)
+                return false;
+
+            if(pArg_->xyLengths_[Rank - 1] % XSrcVectorSize != 0 ||
+               pArg_->xyLengths_[Rank - 1] % YDstVectorSize != 0)
+                return false;
+        };
+
+        if(pArg_->bnScaleStrides_[NumInvariantDim - 1] != 1 && ScaleSrcVectorSize != 1)
+            return false;
+        if(pArg_->bnBiasStrides_[NumInvariantDim - 1] != 1 && BiasSrcVectorSize != 1)
+            return false;
+
+        if(pArg_->bnScaleBiasMeanVarLengths_[NumInvariantDim - 1] % ScaleSrcVectorSize != 0)
+            return false;
+        if(pArg_->bnScaleBiasMeanVarLengths_[NumInvariantDim - 1] % BiasSrcVectorSize != 0)
+            return false;
+
+        if(pArg_->bnMeanVarStrides_[NumInvariantDim - 1] != 1 && MeanVarSrcDstVectorSize != 1)
+            return false;
+
+        if(pArg_->bnScaleBiasMeanVarLengths_[NumInvariantDim - 1] % MeanVarSrcDstVectorSize != 0)
+            return false;
+
+        bool is_valid = true;
+
+        static_for<0, NumInvariantDim, 1>{}([&](auto I) {
+            if(pArg_->xyLengths_[I] != pArg_->bnScaleBiasMeanVarLengths_[I])
+                is_valid = false;
+        });
+
+        if(!is_valid)
+            return false;
+
+        return true;
+    };
+
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(
+        const std::array<index_t, Rank> xyLengths,
+        const std::array<index_t, Rank> xStrides,
+        const std::array<index_t, Rank> yStrides,
+        const std::array<int, NumBatchNormReduceDim> reduceDims,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleBiasMeanVarLengths,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleStrides,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnBiasStrides,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnMeanVarStrides,
+        const void* p_x,
+        const void* p_scale,
+        const void* p_bias,
+        double epsilon,
+        const YElementwiseOp y_elementwise_op,
+        void* p_y,
+        void* resultSaveMean,
+        void* resultSaveInvVariance,
+        double averageFactor,
+        void* resultRunningMean,
+        void* resultRunningVariance) override
+    {
+        return std::make_unique<Argument>(xyLengths,
+                                          xStrides,
+                                          yStrides,
+                                          reduceDims,
+                                          bnScaleBiasMeanVarLengths,
+                                          bnScaleStrides,
+                                          bnBiasStrides,
+                                          bnMeanVarStrides,
+                                          static_cast<const XDataType*>(p_x),
+                                          static_cast<const ScaleDataType*>(p_scale),
+                                          static_cast<const BiasDataType*>(p_bias),
+                                          y_elementwise_op,
+                                          epsilon,
+                                          static_cast<YDataType*>(p_y),
+                                          static_cast<MeanVarDataType*>(resultSaveMean),
+                                          static_cast<MeanVarDataType*>(resultSaveInvVariance),
+                                          averageFactor,
+                                          static_cast<MeanVarDataType*>(resultRunningMean),
+                                          static_cast<MeanVarDataType*>(resultRunningVariance));
+    };
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>();
+    };
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceBatchNormFwdImpl<" << BlockSize << ",";
+        str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
+        str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
+        str << "XSrcYDstVectorDim_" << XSrcYDstVectorDim  << ",";
+        str << "VectorSize_X" << XSrcVectorSize << "_scale_" << ScaleSrcVectorSize << "_bias_" << BiasSrcVectorSize << "_mean_var_" << MeanVarSrcDstVectorSize << "_Y" << YDstVectorSize << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
index 29978458bb202297868af555b3ffecb617e48fb7..9ee6c6f46bad3940515fef1c0cd391667672952d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -118,277 +118,11 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         return PadDescriptor_M_1d(desc_m, gridSize, blockSize);
     }
 
-    static auto MakeAGridDescriptor_AK0_M_AK1(index_t MRaw, index_t KRaw, index_t StrideA)
-    {
-        const auto a_grid_desc_mraw_kraw = [&]() {
-            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both M and K
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(M)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad M, but not K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_right_pad_transform(MRaw, MPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad K, but not M
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k = transform_tensor_descriptor(
-                a_grid_desc_mraw_kraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else
-        {
-            // not pad M or K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-    }
-
-    static auto MakeBGridDescriptor_BK0_N_BK1(index_t KRaw, index_t NRaw, index_t StrideB)
-    {
-        const auto b_grid_desc_nraw_kraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(I1, StrideB));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(StrideB, I1));
-            }
-        }();
-
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto NPad = N - NRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both N and K
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_right_pad_transform(NRaw, NPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(N)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad N, but not K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_right_pad_transform(NRaw, NPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad K, but not N
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k = transform_tensor_descriptor(
-                b_grid_desc_nraw_kraw,
-                make_tuple(make_pass_through_transform(NRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else
-        {
-            // not pad N or K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideC)
-    {
-        const auto c_grid_desc_mraw_nraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(I1, StrideC));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto NPad = N - NRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad M and N
-            return transform_tensor_descriptor(c_grid_desc_mraw_nraw,
-                                               make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                          make_right_pad_transform(NRaw, NPad)),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad M, but not N
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_right_pad_transform(MRaw, MPad), make_pass_through_transform(NRaw)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad N, but not M
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(NRaw, NPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-            // not pad M or N
-            return c_grid_desc_mraw_nraw;
-        }
-    }
-
-    using AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1(1, 1, 1));
-    using BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1(1, 1, 1));
-    using CGridDesc_M_N       = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-    using CGridDesc_M         = decltype(MakeDescriptor_M({1, 1}, {1, 1}, 1, 1));
-
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1<
+        ALayout,
+        BLayout,
+        CLayout,
         ADataType, // TODO: distinguish A/B datatype
         GemmAccDataType,
         CShuffleDataType,
@@ -396,10 +130,8 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_AK0_M_AK1,
-        BGridDesc_BK0_N_BK1,
-        CGridDesc_M_N,
         NumGemmKPrefetchStage,
         BlockSize,
         MPerBlock,
@@ -433,108 +165,82 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         CShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched>;
 
+    using CGridDesc_M = decltype(MakeDescriptor_M({1, 1}, {1, 1}, 1, 1));
+
     // Argument
-    struct Argument : public BaseArgument
+    struct Argument : public tensor_operation::device::BaseArgument, public GridwiseGemm::Problem
     {
-        Argument(const ADataType* p_a_grid_real,
-                 const ADataType* p_a_grid_imag,
-                 const BDataType* p_b_grid_real,
-                 const BDataType* p_b_grid_imag,
-                 CDataType* p_c_grid_real,
-                 CDataType* p_c_grid_imag,
+        using Problem = typename GridwiseGemm::Problem;
+
+        Argument(const ADataType* p_a_grid_real_,
+                 const ADataType* p_a_grid_imag_,
+                 const BDataType* p_b_grid_real_,
+                 const BDataType* p_b_grid_imag_,
+                 CDataType* p_c_grid_real_,
+                 CDataType* p_c_grid_imag_,
                  CDataType* p_workspace,
-                 index_t MRaw,
-                 index_t NRaw,
-                 index_t KRaw,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_real_{p_a_grid_real},
-              p_a_grid_imag_{p_a_grid_imag},
-              p_b_grid_real_{p_b_grid_real},
-              p_b_grid_imag_{p_b_grid_imag},
-              p_c_grid_real_{p_c_grid_real},
-              p_c_grid_imag_{p_c_grid_imag},
-              p_aux_grid_{p_workspace},
-              a_grid_desc_ak0_m_ak1_{DeviceOp::MakeAGridDescriptor_AK0_M_AK1(MRaw, KRaw, StrideA)},
-              b_grid_desc_bk0_n_bk1_{DeviceOp::MakeBGridDescriptor_BK0_N_BK1(KRaw, NRaw, StrideB)},
-              c_grid_desc_m_n_{DeviceOp::MakeCGridDescriptor_M_N(MRaw, NRaw, StrideC)},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_ctile_map_{GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_)},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid_real{p_a_grid_real_},
+              p_a_grid_imag{p_a_grid_imag_},
+              p_b_grid_real{p_b_grid_real_},
+              p_b_grid_imag{p_b_grid_imag_},
+              p_c_grid_real{p_c_grid_real_},
+              p_c_grid_imag{p_c_grid_imag_},
+              p_aux_grid{p_workspace}
         {
-            if(GridwiseGemm::CheckValidity(a_grid_desc_ak0_m_ak1_,
-                                           b_grid_desc_bk0_n_bk1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        c_grid_desc_m_n_);
-            }
-
-            const index_t grid_size = block_2_ctile_map_.CalculateGridSize(c_grid_desc_m_n_);
+            const index_t grid_size = std::get<1>(GridwiseGemm::CalculateGridSize(M_, N_));
 
             if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
             {
-                c_grid_desc_m_ =
-                    DeviceOp::MakeDescriptor_M({MRaw, NRaw}, {StrideC, I1}, grid_size, BlockSize);
+                c_grid_desc_m =
+                    DeviceOp::MakeDescriptor_M({M_, N_}, {StrideC_, I1}, grid_size, BlockSize);
             }
             else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
             {
-                c_grid_desc_m_ =
-                    DeviceOp::MakeDescriptor_M({MRaw, NRaw}, {I1, StrideC}, grid_size, BlockSize);
+                c_grid_desc_m =
+                    DeviceOp::MakeDescriptor_M({M_, N_}, {I1, StrideC_}, grid_size, BlockSize);
             }
 
-            p_aux_2_grid_ = p_workspace + c_grid_desc_m_n_.GetElementSpaceSize();
+            p_aux_2_grid = p_workspace + GetCElementSpaceSize(M_, N_, StrideC_);
         }
 
         //  private:
-        const ADataType* p_a_grid_real_;
-        const ADataType* p_a_grid_imag_;
-        const BDataType* p_b_grid_real_;
-        const BDataType* p_b_grid_imag_;
-        CDataType* p_c_grid_real_;
-        CDataType* p_c_grid_imag_;
-        CDataType* p_aux_grid_;
-        CDataType* p_aux_2_grid_;
-        AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
-        BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        CGridDesc_M c_grid_desc_m_;
-        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            c_grid_desc_mblock_mperblock_nblock_nperblock_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
+        const ADataType* p_a_grid_real;
+        const ADataType* p_a_grid_imag;
+        const BDataType* p_b_grid_real;
+        const BDataType* p_b_grid_imag;
+        CDataType* p_c_grid_real;
+        CDataType* p_c_grid_imag;
+        CDataType* p_aux_grid;
+        CDataType* p_aux_2_grid;
+        CGridDesc_M c_grid_desc_m;
     };
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceOp::Argument;
-
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                            arg.b_grid_desc_bk0_n_bk1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(stream_config.log_level_ > 0)
+            {
+                arg.Print();
+            }
+
+            if(!GridwiseGemm::CheckValidity(arg))
             {
                 throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
+            index_t gdx, gdy, gdz;
+            std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(arg.M, arg.N);
 
-            const auto K =
-                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
+            const auto K = GridwiseGemm::CalculateAK0(arg.K) * AK1;
 
             float ave_time = 0;
 
@@ -578,224 +284,148 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    true>;
-
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
-
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                const auto kernel =
+                    kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, ADataType, CDataType, true>;
+
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_grid,
+                                                   arg);
+
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_real = aux - aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     subtract_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_real_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_real),
                     Subtract{});
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
-
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_grid,
+                                                   arg);
+
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_imag = aux + aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     add_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_imag_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_imag),
                     Add{});
             }
             else
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    false>;
-
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
-
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                const auto kernel =
+                    kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, ADataType, CDataType, false>;
+
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_grid,
+                                                   arg);
+
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_real = aux - aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     subtract_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_real_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_real),
                     Subtract{});
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
-
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_grid,
+                                                   arg);
+
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_imag = aux + aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     add_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_imag_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_imag),
                     Add{});
             }
 
@@ -818,10 +448,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(arg);
     }
 
     // polymorphic
@@ -837,15 +464,15 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                              CDataType* p_c_real,
                              CDataType* p_c_imag,
                              CDataType* p_workspace,
-                             index_t MRaw,
-                             index_t NRaw,
-                             index_t KRaw,
+                             index_t M,
+                             index_t N,
+                             index_t K,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation)
     {
         return Argument{p_a_real,
                         p_a_imag,
@@ -854,15 +481,12 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                         p_c_real,
                         p_c_imag,
                         p_workspace,
-                        MRaw,
-                        NRaw,
-                        KRaw,
+                        M,
+                        N,
+                        K,
                         StrideA,
                         StrideB,
-                        StrideC,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+                        StrideC};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -875,15 +499,15 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                                                       void* p_c_real,
                                                       void* p_c_imag,
                                                       void* p_workspace,
-                                                      index_t MRaw,
-                                                      index_t NRaw,
-                                                      index_t KRaw,
+                                                      index_t M,
+                                                      index_t N,
+                                                      index_t K,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op,
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation,
                                                       index_t /* KBatch */ = 1) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a_real),
@@ -893,15 +517,12 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                                           static_cast<CDataType*>(p_c_real),
                                           static_cast<CDataType*>(p_c_imag),
                                           static_cast<CDataType*>(p_workspace),
-                                          MRaw,
-                                          NRaw,
-                                          KRaw,
+                                          M,
+                                          N,
+                                          K,
                                           StrideA,
                                           StrideB,
-                                          StrideC,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          StrideC);
     }
 
     // polymorphic
@@ -930,16 +551,22 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         return str.str();
     }
 
-    std::size_t GetWorkspaceSize(index_t MRaw,
-                                 index_t NRaw,
-                                 [[maybe_unused]] index_t KRaw,
+    static std::size_t GetCElementSpaceSize(index_t M, index_t N, index_t StrideC)
+    {
+        const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(
+            M, GridwiseGemm::CalculateMPadded(M), N, GridwiseGemm::CalculateNPadded(N), StrideC);
+
+        return c_grid_desc_m_n.GetElementSpaceSize();
+    }
+
+    std::size_t GetWorkspaceSize(index_t M,
+                                 index_t N,
+                                 [[maybe_unused]] index_t K,
                                  [[maybe_unused]] index_t StrideA,
                                  [[maybe_unused]] index_t StrideB,
                                  index_t StrideC) override
     {
-        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(MRaw, NRaw, StrideC);
-
-        return 2 * sizeof(CDataType) * c_grid_desc_m_n.GetElementSpaceSize();
+        return 2 * sizeof(CDataType) * GetCElementSpaceSize(M, N, StrideC);
     }
 };
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index 72c6d0b6f74aab5ec8bd4d0c0d0106fc0ca7bd19..dd57c90898e7c908cb6952e74a2a0decca28a5b5 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -52,7 +52,8 @@ __global__ void
                 e_grid_desc_mblock_mperblock_nblock_nperblock,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -581,7 +582,14 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
+        {
+            return false;
+        }
+
+        if(ck::get_device_name() != "gfx90a" && std::is_same<ADataType, double>::value)
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
index 9f9967c96c285ab3fa9a249630e9878a2f099e82..2ab09ba5c833a489ea22a3f3db06134037bdcedd 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -777,7 +777,19 @@ struct DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
index 806b0c5925631e3f2bfb06aadef144fb307f71c8..eb4db6f8cb098915654ecf32e9ea3c8bc1939ef0 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -379,9 +379,6 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
         InElementwiseOperation,
         WeiElementwiseOperation,
         OutElementwiseOperation,
@@ -428,20 +425,10 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                  std::vector<ck::index_t> conv_filter_strides,
                  std::vector<ck::index_t> conv_filter_dilations,
                  std::vector<ck::index_t> input_left_pads,
-                 std::vector<ck::index_t> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
-                 InElementwiseOperation in_element_op,
-                 WeiElementwiseOperation wei_element_op,
-                 OutElementwiseOperation out_element_op)
+                 std::vector<ck::index_t> input_right_pads)
             : p_a_grid_{p_out_grid},
               p_b_grid_{p_wei_grid},
               p_c_grid_{p_in_grid},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{out_element_op},
-              b_element_op_{wei_element_op},
-              c_element_op_{in_element_op},
               Conv_N_{N},
               Conv_K_{K},
               Conv_C_{C},
@@ -495,18 +482,6 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                     a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                     b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                     c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                    auto block_2_ctile_map =
-                        GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01, N01);
-
-                    if(GridwiseGemm::CheckValidity(
-                           descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                    {
-                        c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                            GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(descs[I2]));
-
-                        block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                    }
                 }
             }
         }
@@ -517,14 +492,6 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         std::vector<AGridDesc_K0_M_K1> a_grid_desc_k0_m_k1_container_;
         std::vector<BGridDesc_K0_N_K1> b_grid_desc_k0_n_k1_container_;
         std::vector<CGridDesc_M_N> c_grid_desc_m_n_container_;
-        std::vector<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_;
-        std::vector<typename GridwiseGemm::DefaultBlock2CTileMap> block_2_ctile_map_container_;
-        index_t M01_;
-        index_t N01_;
-        OutElementwiseOperation a_element_op_;
-        WeiElementwiseOperation b_element_op_;
-        InElementwiseOperation c_element_op_;
         // for checking IsSupportedArgument()
         index_t Conv_N_;
         index_t Conv_K_;
@@ -567,103 +534,68 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I0) << ", "
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I1) << "}"
                               << std::endl;
-
-                    std::cout << "arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I0)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I1)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I2)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I3)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I4)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I5)
-                              << " ) " << std::endl;
                 }
 #endif
 
                 if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                                 arg.b_grid_desc_k0_n_k1_container_[i],
-                                                arg.c_grid_desc_m_n_container_[i],
-                                                arg.block_2_ctile_map_container_[i]))
+                                                arg.c_grid_desc_m_n_container_[i]))
                 {
                     throw std::runtime_error(
                         "wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v3r1 has invalid setting");
                 }
 
-                const index_t grid_size = arg.block_2_ctile_map_container_[i].CalculateGridSize(
-                    arg.c_grid_desc_m_n_container_[i]);
+                const auto [gdx, gdy, gdz] =
+                    GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_container_[i]);
 
                 const auto K = arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I0) *
                                arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I2);
 
                 if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        true>;
-
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                true>;
+
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        false>;
-
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                false>;
+
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
             }
             return ave_time;
@@ -716,8 +648,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         {
             if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                             arg.b_grid_desc_k0_n_k1_container_[i],
-                                            arg.c_grid_desc_m_n_container_[i],
-                                            arg.block_2_ctile_map_container_[i]))
+                                            arg.c_grid_desc_m_n_container_[i]))
             {
                 return false;
             }
@@ -742,10 +673,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                              std::vector<ck::index_t> conv_filter_strides,
                              std::vector<ck::index_t> conv_filter_dilations,
                              std::vector<ck::index_t> input_left_pads,
-                             std::vector<ck::index_t> input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op)
+                             std::vector<ck::index_t> input_right_pads)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
@@ -759,12 +687,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
-                        input_right_pads,
-                        1,
-                        1,
-                        in_element_op,
-                        wei_element_op,
-                        out_element_op};
+                        input_right_pads};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -783,9 +706,9 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         std::vector<ck::index_t> conv_filter_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
-                        InElementwiseOperation in_element_op,
-                        WeiElementwiseOperation wei_element_op,
-                        OutElementwiseOperation out_element_op) override
+                        InElementwiseOperation,
+                        WeiElementwiseOperation,
+                        OutElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<InDataType*>(p_in_grid),
                                           static_cast<const WeiDataType*>(p_wei_grid),
@@ -799,12 +722,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
-                                          input_right_pads,
-                                          1,
-                                          1,
-                                          in_element_op,
-                                          wei_element_op,
-                                          out_element_op);
+                                          input_right_pads);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
@@ -822,7 +740,14 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
index ff49d3b82e9ae717aef0a0c8fe91e661d8233f57..bb8f5316134d8941fc24d17ed9a65d280353b9f8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -956,7 +956,19 @@ struct
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
index 4934599ee48fbf8321b43131c05087bd5dcb7b5e..1bd1e553cd36022ff86ba37aee57157e0feca1e1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -913,7 +913,19 @@ struct DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
index 731dfc5ea1f92160feafbe2cd710eebaf168380b..de6bf27fec37511ae40d98e6dda7b4352e1fa139 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -880,7 +880,17 @@ struct DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_W
             << MPerBlock << ", "
             << NPerBlock << ", "
             << K0PerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << K1 << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
index 5848006931204d8ffc758f12253d4c13385f9482..88615bba3134311b4120f4aa4483050df3e084a8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -329,9 +329,6 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
         InElementwiseOperation,
         WeiElementwiseOperation,
         OutElementwiseOperation,
@@ -378,25 +375,13 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                  std::vector<ck::index_t> conv_filter_strides,
                  std::vector<ck::index_t> conv_filter_dilations,
                  std::vector<ck::index_t> input_left_pads,
-                 std::vector<ck::index_t> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
-                 InElementwiseOperation in_element_op,
-                 WeiElementwiseOperation wei_element_op,
-                 OutElementwiseOperation out_element_op)
+                 std::vector<ck::index_t> input_right_pads)
             : p_a_grid_{p_in_grid},
               p_b_grid_{p_wei_grid},
               p_c_grid_{p_out_grid},
               a_grid_desc_k0_m_k1_{},
               b_grid_desc_k0_n_k1_{},
               c_grid_desc_m_n_{},
-              c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
-              block_2_ctile_map_{},
-              M01_{M01},
-              N01_{N01},
-              in_element_op_{in_element_op},
-              wei_element_op_{wei_element_op},
-              out_element_op_{out_element_op},
               Conv_N_{N},
               Conv_K_{K},
               Conv_C_{C},
@@ -420,17 +405,6 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
             a_grid_desc_k0_m_k1_ = descs[I0];
             b_grid_desc_k0_n_k1_ = descs[I1];
             c_grid_desc_m_n_     = descs[I2];
-            block_2_ctile_map_ =
-                GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_, M01, N01);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_k0_m_k1_,
-                                           b_grid_desc_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
-                    GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
-            }
         }
 
         //  private:
@@ -440,14 +414,6 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
         BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
         CGridDesc_M_N c_grid_desc_m_n_;
-        typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        InElementwiseOperation in_element_op_;
-        WeiElementwiseOperation wei_element_op_;
-        OutElementwiseOperation out_element_op_;
         // for checking IsSupportedArgument()
         index_t Conv_N_;
         index_t Conv_K_;
@@ -479,17 +445,14 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                           << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
             }
 #endif
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                            arg.b_grid_desc_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(
+                   arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_))
             {
                 throw std::runtime_error(
                     "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3 has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
+            const auto [gdx, gdy, gdz] = GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_);
 
             const auto K =
                 arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
@@ -498,22 +461,18 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    InElementwiseOperation,
-                    WeiElementwiseOperation,
-                    OutElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    true>;
+                const auto kernel =
+                    kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                            ADataType, // TODO: distiguish A/B datatype
+                                            CDataType,
+                                            DeviceOp::AGridDesc_K0_M_K1,
+                                            DeviceOp::BGridDesc_K0_N_K1,
+                                            DeviceOp::CGridDesc_M_N,
+                                            true>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
-                                                  dim3(grid_size),
+                                                  dim3(gdx, gdy, gdz),
                                                   dim3(BlockSize),
                                                   0,
                                                   arg.p_a_grid_,
@@ -521,30 +480,22 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                                   arg.p_c_grid_,
                                                   arg.a_grid_desc_k0_m_k1_,
                                                   arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.in_element_op_,
-                                                  arg.wei_element_op_,
-                                                  arg.out_element_op_,
-                                                  arg.block_2_ctile_map_);
+                                                  arg.c_grid_desc_m_n_);
             }
             else
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    InElementwiseOperation,
-                    WeiElementwiseOperation,
-                    OutElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    false>;
+                const auto kernel =
+                    kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                            ADataType, // TODO: distiguish A/B datatype
+                                            CDataType,
+                                            DeviceOp::AGridDesc_K0_M_K1,
+                                            DeviceOp::BGridDesc_K0_N_K1,
+                                            DeviceOp::CGridDesc_M_N,
+                                            false>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
-                                                  dim3(grid_size),
+                                                  dim3(gdx, gdy, gdz),
                                                   dim3(BlockSize),
                                                   0,
                                                   arg.p_a_grid_,
@@ -552,11 +503,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                                   arg.p_c_grid_,
                                                   arg.a_grid_desc_k0_m_k1_,
                                                   arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.in_element_op_,
-                                                  arg.wei_element_op_,
-                                                  arg.out_element_op_,
-                                                  arg.block_2_ctile_map_);
+                                                  arg.c_grid_desc_m_n_);
             }
 
             return ave_time;
@@ -616,10 +563,8 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         }
 
         // Gridwise GEMM size
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                           arg.b_grid_desc_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(
+            arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);
     }
 
     bool IsSupportedArgument(const BaseArgument* p_arg) override
@@ -639,10 +584,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                              std::vector<ck::index_t> conv_filter_strides,
                              std::vector<ck::index_t> conv_filter_dilations,
                              std::vector<ck::index_t> input_left_pads,
-                             std::vector<ck::index_t> input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op)
+                             std::vector<ck::index_t> input_right_pads)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
@@ -656,12 +598,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
-                        input_right_pads,
-                        1,
-                        1,
-                        in_element_op,
-                        wei_element_op,
-                        out_element_op};
+                        input_right_pads};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -680,9 +617,9 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         std::vector<ck::index_t> conv_filter_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
-                        InElementwiseOperation in_element_op,
-                        WeiElementwiseOperation wei_element_op,
-                        OutElementwiseOperation out_element_op) override
+                        InElementwiseOperation,
+                        WeiElementwiseOperation,
+                        OutElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_grid),
                                           static_cast<const WeiDataType*>(p_wei_grid),
@@ -696,12 +633,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
-                                          input_right_pads,
-                                          1,
-                                          1,
-                                          in_element_op,
-                                          wei_element_op,
-                                          out_element_op);
+                                          input_right_pads);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
@@ -720,7 +652,16 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
             << MPerBlock << ", "
             << NPerBlock << ", "
             << K0PerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp
index f69d8f18ae036b8e7cd060a80a0cca12bfc2a050..cd89f3232cb94e9ce9a68c06587758ac9574816f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef DEVICE_CONV3D_FWD_NAIVE_HPP
 #define DEVICE_CONV3D_FWD_NAIVE_HPP
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
index fe2e6c30890ccaa59f9f42ff8620f20485d059cf..bc6582835e051c8335b7c4e12c5492b436a34392 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef DEVICE_CONV3D_FWD_XDL_HPP
 #define DEVICE_CONV3D_FWD_XDL_HPP
@@ -55,7 +55,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / num_batches);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -630,7 +631,19 @@ struct DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
index 5bef0e2a3eff07226ab5d61279a63a8edea6feb6..3178f73f4b530e67b399cd0ae71064a4b66fa837 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -1393,7 +1393,9 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Dl
     static bool IsSupportedArgument(const Argument& arg)
     {
         // check device
-        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030"))
+        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+             ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+             ck::get_device_name() == "gfx1102"))
         {
             return false;
         }
@@ -1567,7 +1569,8 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Dl
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1
             << ">";
         if constexpr(ConvBackwardDataSpecialization ==
                      ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0){
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
index 7951713938f2ca3dbffb9d990a98e0408e1c722e..77ad61d7ec5c446eed5c2171e36f2af2490d51a2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -980,9 +980,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
         InElementwiseOperation,
         WeiElementwiseOperation,
         OutElementwiseOperation,
@@ -1029,20 +1026,10 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                  std::vector<ck::index_t> conv_filter_strides,
                  std::vector<ck::index_t> conv_filter_dilations,
                  std::vector<ck::index_t> input_left_pads,
-                 std::vector<ck::index_t> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
-                 InElementwiseOperation in_element_op,
-                 WeiElementwiseOperation wei_element_op,
-                 OutElementwiseOperation out_element_op)
+                 std::vector<ck::index_t> input_right_pads)
             : p_a_grid_{p_out_grid},
               p_b_grid_{p_wei_grid},
               p_c_grid_{p_in_grid},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{out_element_op},
-              b_element_op_{wei_element_op},
-              c_element_op_{in_element_op},
               Conv_N_{N},
               Conv_K_{K},
               Conv_C_{C},
@@ -1092,17 +1079,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                 a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                 b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                 c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                auto block_2_ctile_map =
-                    GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01_, N01_);
-
-                if(GridwiseGemm::CheckValidity(descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                {
-                    c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                        GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(descs[I2]));
-
-                    block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                }
             }
         }
         template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
@@ -1150,18 +1126,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                     a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                     b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                     c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                    auto block_2_ctile_map =
-                        GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01_, N01_);
-
-                    if(GridwiseGemm::CheckValidity(
-                           descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                    {
-                        c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                            GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(descs[I2]));
-
-                        block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                    }
                 }
             }
         }
@@ -1218,19 +1182,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                         a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                         b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                         c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                        auto block_2_ctile_map =
-                            GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01_, N01_);
-
-                        if(GridwiseGemm::CheckValidity(
-                               descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                        {
-                            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                                GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(
-                                    descs[I2]));
-
-                            block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                        }
                     }
                 }
             }
@@ -1242,11 +1193,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
         std::vector<AGridDesc_K0_M_K1> a_grid_desc_k0_m_k1_container_;
         std::vector<BGridDesc_K0_N_K1> b_grid_desc_k0_n_k1_container_;
         std::vector<CGridDesc_M_N> c_grid_desc_m_n_container_;
-        std::vector<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_;
-        std::vector<typename GridwiseGemm::DefaultBlock2CTileMap> block_2_ctile_map_container_;
-        index_t M01_;
-        index_t N01_;
         OutElementwiseOperation a_element_op_;
         WeiElementwiseOperation b_element_op_;
         InElementwiseOperation c_element_op_;
@@ -1276,123 +1222,84 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
             {
 #if DEBUG_LOG
                 {
-                    std::cout << "arg.a_grid_desc_k0_m_k1_container_{"
+                    std::cout << "arg.a_grid_desc_k0_m_k1{"
                               << arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I0) << ", "
                               << arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I1) << ", "
                               << arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I2) << "}"
                               << std::endl;
 
-                    std::cout << "arg.b_grid_desc_k0_n_k1_container_{"
+                    std::cout << "arg.b_grid_desc_k0_n_k1{"
                               << arg.b_grid_desc_k0_n_k1_container_[i].GetLength(I0) << ", "
                               << arg.b_grid_desc_k0_n_k1_container_[i].GetLength(I1) << ", "
                               << arg.b_grid_desc_k0_n_k1_container_[i].GetLength(I2) << "}"
                               << std::endl;
 
-                    std::cout << "arg.c_grid_desc_m_n_container_{ "
+                    std::cout << "arg.c_grid_desc_m_n{"
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I0) << ", "
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I1) << "}"
                               << std::endl;
-
-                    std::cout << "arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I0)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I1)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I2)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I3)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I4)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I5)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I6)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I7)
-                              << " ) " << std::endl;
                 }
 #endif
 
                 if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                                 arg.b_grid_desc_k0_n_k1_container_[i],
-                                                arg.c_grid_desc_m_n_container_[i],
-                                                arg.block_2_ctile_map_container_[i]))
+                                                arg.c_grid_desc_m_n_container_[i]))
                 {
                     throw std::runtime_error(
-                        "wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v3r1 has invalid setting");
+                        "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3 has invalid setting");
                 }
 
-                const index_t grid_size = arg.block_2_ctile_map_container_[i].CalculateGridSize(
-                    arg.c_grid_desc_m_n_container_[i]);
+                const auto [gdx, gdy, gdz] =
+                    GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_container_[i]);
 
                 const auto K = arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I0) *
                                arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I2);
 
                 if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        true>;
-
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                true>;
+
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        false>;
-
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                false>;
+
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
             }
             return ave_time;
@@ -1446,8 +1353,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
         {
             if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                             arg.b_grid_desc_k0_n_k1_container_[i],
-                                            arg.c_grid_desc_m_n_container_[i],
-                                            arg.block_2_ctile_map_container_[i]))
+                                            arg.c_grid_desc_m_n_container_[i]))
             {
                 return false;
             }
@@ -1472,10 +1378,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                              std::vector<ck::index_t> conv_filter_strides,
                              std::vector<ck::index_t> conv_filter_dilations,
                              std::vector<ck::index_t> input_left_pads,
-                             std::vector<ck::index_t> input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op)
+                             std::vector<ck::index_t> input_right_pads)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
@@ -1489,12 +1392,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
-                        input_right_pads,
-                        1,
-                        1,
-                        in_element_op,
-                        wei_element_op,
-                        out_element_op};
+                        input_right_pads};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -1513,9 +1411,9 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                         std::vector<ck::index_t> conv_filter_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
-                        InElementwiseOperation in_element_op,
-                        WeiElementwiseOperation wei_element_op,
-                        OutElementwiseOperation out_element_op) override
+                        InElementwiseOperation,
+                        WeiElementwiseOperation,
+                        OutElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<InDataType*>(p_in_grid),
                                           static_cast<const WeiDataType*>(p_wei_grid),
@@ -1529,12 +1427,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
-                                          input_right_pads,
-                                          1,
-                                          1,
-                                          in_element_op,
-                                          wei_element_op,
-                                          out_element_op);
+                                          input_right_pads);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
@@ -1552,7 +1445,14 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">";
         if constexpr(ConvBackwardDataSpecialization ==
                      ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0){
diff --git a/include/ck/tensor_operation/gpu/device/device_elementwise_2d.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
similarity index 85%
rename from include/ck/tensor_operation/gpu/device/device_elementwise_2d.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
index 23aada0f448c81c83ea2d06071161ff81f3b4206..02ef29e32ddc2bbd985fb4f46c6eb3825c00c11a 100644
--- a/include/ck/tensor_operation/gpu/device/device_elementwise_2d.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -8,11 +8,12 @@
 
 #include "ck/utility/math.hpp"
 #include "ck/utility/sequence.hpp"
-#include "ck/tensor_operation/gpu/device/device_elementwise_base.hpp"
+#include "ck/tensor_operation/gpu/device/device_elementwise.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 
 #include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -26,10 +27,10 @@ template <typename InDataTypeTuple,
           index_t NPerThread,
           typename InScalarPerVectorSeq,
           typename OutScalarPerVectorSeq>
-struct DeviceElementwise : public DeviceElementwiseBase<InDataTypeTuple,
-                                                        OutDataTypeTuple,
-                                                        ElementwiseOperation,
-                                                        NumDim_m + NumDim_n>
+struct DeviceElementwise2dImpl : public DeviceElementwise<InDataTypeTuple,
+                                                          OutDataTypeTuple,
+                                                          ElementwiseOperation,
+                                                          NumDim_m + NumDim_n>
 {
     static constexpr index_t NumDim = NumDim_m + NumDim_n;
 
@@ -171,10 +172,7 @@ struct DeviceElementwise : public DeviceElementwiseBase<InDataTypeTuple,
               inStridesArray_(inStridesArray),
               outStridesArray_(outStridesArray),
               elementwise_op_(elementwise_op),
-              blockSize_(256),
-              gridSize_(120), // FIXME - Calculate the grid size by number of CU in the future
-              num_threads_m_((gridSize_ * blockSize_) / 16),
-              num_threads_n_(16)
+              blockSize_(256)
         {
             static_assert(NumDim_m > 0, "");
             static_assert(NumDim_n > 0, "");
@@ -192,34 +190,10 @@ struct DeviceElementwise : public DeviceElementwiseBase<InDataTypeTuple,
                     return static_cast<DataType*>(out_dev_buffers[I.value]);
                 },
                 Number<NumOutput>{});
-
-            in_grid_2d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_MN(lengths,
-                                             inStridesArray[I.value],
-                                             gridSize_,
-                                             blockSize_,
-                                             num_threads_m_,
-                                             num_threads_n_);
-                },
-                Number<NumInput>{});
-
-            out_grid_2d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_MN(lengths,
-                                             outStridesArray[I.value],
-                                             gridSize_,
-                                             blockSize_,
-                                             num_threads_m_,
-                                             num_threads_n_);
-                },
-                Number<NumOutput>{});
         }
 
         InDataTypePointerTuple in_dev_buffers_;
         OutDataTypePointerTuple out_dev_buffers_;
-        InGrid2dDescTuple in_grid_2d_desc_tuple_;
-        OutGrid2dDescTuple out_grid_2d_desc_tuple_;
 
         std::array<index_t, NumDim> lengths_;
         std::array<std::array<index_t, NumDim>, NumInput> inStridesArray_;
@@ -227,15 +201,38 @@ struct DeviceElementwise : public DeviceElementwiseBase<InDataTypeTuple,
 
         ElementwiseOperation elementwise_op_;
         index_t blockSize_;
-        index_t gridSize_;
-        index_t num_threads_m_;
-        index_t num_threads_n_;
     };
 
     struct Invoker : public BaseInvoker
     {
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
+            index_t gridSize      = getAvailableComputeUnitCount(stream_config);
+            index_t num_threads_m = (gridSize * arg.blockSize_) / 16;
+            index_t num_threads_n = 16;
+
+            auto in_grid_2d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_MN(arg.lengths_,
+                                             arg.inStridesArray_[I.value],
+                                             gridSize,
+                                             arg.blockSize_,
+                                             num_threads_m,
+                                             num_threads_n);
+                },
+                Number<NumInput>{});
+
+            auto out_grid_2d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_MN(arg.lengths_,
+                                             arg.outStridesArray_[I.value],
+                                             gridSize,
+                                             arg.blockSize_,
+                                             num_threads_m,
+                                             num_threads_n);
+                },
+                Number<NumOutput>{});
+
             const auto kernel = kernel_elementwise_2d<GridwiseElementwise,
                                                       InGrid2dDescTuple,
                                                       OutGrid2dDescTuple,
@@ -245,16 +242,16 @@ struct DeviceElementwise : public DeviceElementwiseBase<InDataTypeTuple,
 
             float elapsed_time = launch_and_time_kernel(stream_config,
                                                         kernel,
-                                                        dim3(arg.gridSize_),
+                                                        dim3(gridSize),
                                                         dim3(arg.blockSize_),
                                                         0,
-                                                        arg.in_grid_2d_desc_tuple_,
-                                                        arg.out_grid_2d_desc_tuple_,
+                                                        in_grid_2d_desc_tuple,
+                                                        out_grid_2d_desc_tuple,
                                                         arg.in_dev_buffers_,
                                                         arg.out_dev_buffers_,
                                                         arg.elementwise_op_,
-                                                        arg.num_threads_m_,
-                                                        arg.num_threads_n_);
+                                                        num_threads_m,
+                                                        num_threads_n);
             return elapsed_time;
         }
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
similarity index 91%
rename from include/ck/tensor_operation/gpu/device/impl/device_elementwise.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
index 8e6288009860ad9fa950310f15e7b0ffcffc7486..e203468c6174f0ae2c8bde33a716d0665f2e8b6c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -8,11 +8,12 @@
 
 #include "ck/utility/math.hpp"
 #include "ck/utility/sequence.hpp"
-#include "ck/tensor_operation/gpu/device/device_elementwise_base.hpp"
+#include "ck/tensor_operation/gpu/device/device_elementwise.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 
 #include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -25,8 +26,8 @@ template <typename InDataTypeTuple,
           index_t MPerThread,
           typename InScalarPerVectorSeq,
           typename OutScalarPerVectorSeq>
-struct DeviceElementwise
-    : public DeviceElementwiseBase<InDataTypeTuple, OutDataTypeTuple, ElementwiseOperation, NumDim>
+struct DeviceElementwiseImpl
+    : public DeviceElementwise<InDataTypeTuple, OutDataTypeTuple, ElementwiseOperation, NumDim>
 {
     static constexpr int NumInput  = InDataTypeTuple::Size();
     static constexpr int NumOutput = OutDataTypeTuple::Size();
@@ -144,8 +145,7 @@ struct DeviceElementwise
               inStridesArray_(inStridesArray),
               outStridesArray_(outStridesArray),
               elementwise_op_(elementwise_op),
-              blockSize_(256),
-              gridSize_(120) // FIXME - Calculate the grid size by number of CU in the future
+              blockSize_(256)
         {
             in_dev_buffers_ = generate_tuple(
                 [&](auto I) {
@@ -160,26 +160,10 @@ struct DeviceElementwise
                     return static_cast<DataType*>(out_dev_buffers[I.value]);
                 },
                 Number<NumOutput>{});
-
-            in_grid_1d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_M(
-                        lengths, inStridesArray[I.value], gridSize_, blockSize_);
-                },
-                Number<NumInput>{});
-
-            out_grid_1d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_M(
-                        lengths, outStridesArray[I.value], gridSize_, blockSize_);
-                },
-                Number<NumOutput>{});
         }
 
         InDataTypePointerTuple in_dev_buffers_;
         OutDataTypePointerTuple out_dev_buffers_;
-        InGrid1dDescTuple in_grid_1d_desc_tuple_;
-        OutGrid1dDescTuple out_grid_1d_desc_tuple_;
 
         std::array<index_t, NumDim> lengths_;
         std::array<std::array<index_t, NumDim>, NumInput> inStridesArray_;
@@ -187,13 +171,28 @@ struct DeviceElementwise
 
         ElementwiseOperation elementwise_op_;
         index_t blockSize_;
-        index_t gridSize_;
     };
 
     struct Invoker : public BaseInvoker
     {
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
+            index_t gridSize = getAvailableComputeUnitCount(stream_config);
+
+            auto in_grid_1d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_M(
+                        arg.lengths_, arg.inStridesArray_[I.value], gridSize, arg.blockSize_);
+                },
+                Number<NumInput>{});
+
+            auto out_grid_1d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_M(
+                        arg.lengths_, arg.outStridesArray_[I.value], gridSize, arg.blockSize_);
+                },
+                Number<NumOutput>{});
+
             const auto kernel = kernel_elementwise_1d<GridwiseElementwise,
                                                       InGrid1dDescTuple,
                                                       OutGrid1dDescTuple,
@@ -203,11 +202,11 @@ struct DeviceElementwise
 
             float elapsed_time = launch_and_time_kernel(stream_config,
                                                         kernel,
-                                                        dim3(arg.gridSize_),
+                                                        dim3(gridSize),
                                                         dim3(arg.blockSize_),
                                                         0,
-                                                        arg.in_grid_1d_desc_tuple_,
-                                                        arg.out_grid_1d_desc_tuple_,
+                                                        in_grid_1d_desc_tuple,
+                                                        out_grid_1d_desc_tuple,
                                                         arg.in_dev_buffers_,
                                                         arg.out_dev_buffers_,
                                                         arg.elementwise_op_);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp
index 8ffc5ef9fb48b85c6f685166a4a842da09e95acf..c3416758d1277d6cc7d69e4aa4115b57af4c9cd1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -270,18 +270,18 @@ struct DeviceElementwiseNormalizationImpl
                  const std::vector<index_t> reduceDims,
                  XElementwiseOperation x_elementwise_op,
                  YElementwiseOperation y_elementwise_op,
-                 AccDataType epsilon,
+                 double epsilon,
                  const std::array<const void*, NumInput> in_dev_buffers,
                  const GammaDataType* p_gamma,
                  const BetaDataType* p_beta,
                  YDataType* p_y)
-            : epsilon_(epsilon),
-              p_gamma_(p_gamma),
+            : p_gamma_(p_gamma),
               p_beta_(p_beta),
               p_y_(p_y),
               x_elementwise_op_(x_elementwise_op),
               y_elementwise_op_(y_elementwise_op)
         {
+            epsilon_ = static_cast<AccDataType>(epsilon);
 
             Lengths_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(lengths, reduceDims);
             for(int i = 0; i < NumInput; i++)
@@ -533,6 +533,11 @@ struct DeviceElementwiseNormalizationImpl
                 return (false);
         }
 
+        if(p_arg_->x_lds_size_ >= 65536)
+        {
+            return (false);
+        }
+
         return true;
     };
 
@@ -543,7 +548,7 @@ struct DeviceElementwiseNormalizationImpl
                         const std::vector<index_t> betaStrides,
                         const std::vector<index_t> yStrides,
                         const std::vector<index_t> reduceDims,
-                        AccDataType epsilon,
+                        double epsilon,
                         const std::array<const void*, NumInput> in_dev_buffers,
                         const void* p_gamma,
                         const void* p_beta,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
index b9a64e8c4b0bde853c1296eb75413110da53fb7e..63f7fa706870ee65cdff0e7db2c70b4447e209bf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -862,7 +862,15 @@ struct DeviceGemmBiasAddReduce_Xdl_CShuffle : public DeviceGemmReduce<1, ReduceO
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
index af1989fc4a2cc41783806bbe3c2f67718e45e524..13e9f96910fadeb0fdda4ef723df3f260bc7ee7b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -485,7 +485,9 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030")
+        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+           ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..ad51096db7f6d8aa1cd8b4a8c7590ff385009943
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
@@ -0,0 +1,673 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemm,
+          typename ABDataType,
+          typename DsPointer,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AGridDesc_K0_M0_M1_K1,
+          typename BGridDesc_K0_N0_N1_K1,
+          typename DsGridDesc_M0_M10_M11_N0_N10_N11,
+          typename CGridDesc_M0_M10_M11_N0_N10_N11,
+          typename Block2CTileMap,
+          bool HasMainKBlockLoop,
+          bool HasDoubleTailKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_dl_multiple_d(
+            const ABDataType* __restrict__ p_a_grid,
+            const ABDataType* __restrict__ p_b_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const AGridDesc_K0_M0_M1_K1 a_grid_desc_k0_m0_m1_k1,
+            const BGridDesc_K0_N0_N1_K1 b_grid_desc_k0_n0_n1_k1,
+            const DsGridDesc_M0_M10_M11_N0_N10_N11 ds_grid_desc_m0_m10_m11_n0_n10_n11,
+            const CGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11,
+            const Block2CTileMap block_2_ctile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
+
+    constexpr index_t shared_block_size =
+        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);
+
+    __shared__ ABDataType p_shared[shared_block_size];
+
+    GridwiseGemm::Run(p_a_grid,
+                      p_b_grid,
+                      p_ds_grid,
+                      p_e_grid,
+                      p_shared,
+                      a_element_op,
+                      b_element_op,
+                      cde_element_op,
+                      a_grid_desc_k0_m0_m1_k1,
+                      b_grid_desc_k0_n0_n1_k1,
+                      ds_grid_desc_m0_m10_m11_n0_n10_n11,
+                      e_grid_desc_m0_m10_m11_n0_n10_n11,
+                      block_2_ctile_map,
+                      integral_constant<bool, HasMainKBlockLoop>{},
+                      integral_constant<bool, HasDoubleTailKBlockLoop>{});
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = a_grid_desc_k0_m0_m1_k1;
+    ignore = b_grid_desc_k0_n0_n1_k1;
+    ignore = ds_grid_desc_m0_m10_m11_n0_n10_n11;
+    ignore = e_grid_desc_m0_m10_m11_n0_n10_n11;
+    ignore = block_2_ctile_map;
+#endif
+}
+} // namespace ck
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t K1,
+          index_t M1PerThread,
+          index_t N1PerThread,
+          index_t KPerThread,
+          typename M1N1ThreadClusterM1Xs,
+          typename M1N1ThreadClusterN1Xs,
+          typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
+          typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+          typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
+          typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector,
+          enable_if_t<
+              is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+                  is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
+              bool> = false>
+struct DeviceGemmMultipleD_Dl : public DeviceGemmMultipleD<ALayout,
+                                                           BLayout,
+                                                           DsLayout,
+                                                           ELayout,
+                                                           ADataType,
+                                                           BDataType,
+                                                           DsDataType,
+                                                           EDataType,
+                                                           AElementwiseOperation,
+                                                           BElementwiseOperation,
+                                                           CDEElementwiseOperation>
+
+{
+    using DeviceOp                      = DeviceGemmMultipleD_Dl;
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr auto K1Number = Number<K1>{};
+
+    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
+    {
+        assert(K % K1 == 0);
+
+        const index_t K0 = K / K1;
+
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(M, PadM)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
+    {
+        assert(K % K1 == 0);
+
+        const index_t K0 = K / K1;
+
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    template <typename ELay>
+    static auto MakeEGridDescriptor_M_N(index_t M, index_t N, index_t StrideE)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideE));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                                         const std::array<index_t, NumDTensor>& NRaws,
+                                         const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return DeviceOp::MakeEGridDescriptor_M_N<DLayout>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
+    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
+    using DsGridDesc_M_N    = decltype(MakeDsGridDescriptor_M_N({}, {}, {}));
+    using EGridDesc_M_N     = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
+
+    // GridwiseGemm
+    using GridwiseGemm =
+        GridwiseGemmDlMultipleD_km_kn_mn<BlockSize,
+                                         ADataType,
+                                         AccDataType,
+                                         DsDataType,
+                                         EDataType,
+                                         AElementwiseOperation,
+                                         BElementwiseOperation,
+                                         CDEElementwiseOperation,
+                                         InMemoryDataOperationEnum::Set,
+                                         AGridDesc_K0_M_K1,
+                                         BGridDesc_K0_N_K1,
+                                         EGridDesc_M_N,
+                                         MPerBlock,
+                                         NPerBlock,
+                                         K0PerBlock,
+                                         K1,
+                                         M1PerThread,
+                                         N1PerThread,
+                                         KPerThread,
+                                         M1N1ThreadClusterM1Xs,
+                                         M1N1ThreadClusterN1Xs,
+                                         ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterArrangeOrder,
+                                         ABlockTransferSrcAccessOrder,
+                                         ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+                                         ABlockTransferSrcVectorTensorContiguousDimOrder,
+                                         ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+                                         BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterArrangeOrder,
+                                         BBlockTransferSrcAccessOrder,
+                                         BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+                                         BBlockTransferSrcVectorTensorContiguousDimOrder,
+                                         BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+                                         CThreadTransferSrcDstAccessOrder,
+                                         CThreadTransferSrcDstVectorDim,
+                                         CThreadTransferDstScalarPerVector>;
+
+    using AGridDesc_K0_M0_M1_K1 =
+        decltype(GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
+    using BGridDesc_K0_N0_N1_K1 =
+        decltype(GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(BGridDesc_K0_N_K1{}));
+    using DsGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(DsGridDesc_M_N{}));
+    using EGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(EGridDesc_M_N{}));
+    using DefaultBlock2CTileMap =
+        decltype(GridwiseGemm::MakeDefaultBlock2CTileMap(EGridDesc_M_N{}));
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a_grid,
+                 const void* p_b_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 index_t M,
+                 index_t N,
+                 index_t K,
+                 index_t StrideA,
+                 index_t StrideB,
+                 std::array<index_t, NumDTensor> StrideDs,
+                 index_t StrideE,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
+              p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              a_grid_desc_k0_m0_m1_k1_{},
+              b_grid_desc_k0_n0_n1_k1_{},
+              e_grid_desc_m0_m10_m11_n0_n10_n11_{},
+              block_2_ctile_map_{},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op}
+        {
+            a_grid_desc_k0_m_k1_ =
+                DeviceGemmMultipleD_Dl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
+            b_grid_desc_k0_n_k1_ =
+                DeviceGemmMultipleD_Dl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    DeviceOp::MakeEGridDescriptor_M_N<DLayout>(M, N, StrideDs[i]);
+            });
+            e_grid_desc_m_n_ =
+                DeviceGemmMultipleD_Dl::MakeEGridDescriptor_M_N<ELayout>(M, N, StrideE);
+
+            if(GridwiseGemm::CheckValidity(
+                   a_grid_desc_k0_m_k1_, b_grid_desc_k0_n_k1_, e_grid_desc_m_n_))
+            {
+                a_grid_desc_k0_m0_m1_k1_ =
+                    GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(a_grid_desc_k0_m_k1_);
+                b_grid_desc_k0_n0_n1_k1_ =
+                    GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(b_grid_desc_k0_n_k1_);
+
+                ds_grid_desc_m0_m10_m11_n0_n10_n11_ =
+                    GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(ds_grid_desc_m_n_);
+
+                e_grid_desc_m0_m10_m11_n0_n10_n11_ =
+                    GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(e_grid_desc_m_n_);
+
+                block_2_ctile_map_ = GridwiseGemm::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_);
+            }
+        }
+
+        //  private:
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseGemm::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        AGridDesc_K0_M0_M1_K1 a_grid_desc_k0_m0_m1_k1_;
+        BGridDesc_K0_N0_N1_K1 b_grid_desc_k0_n0_n1_k1_;
+        DsGridDesc_M0_M10_M11_N0_N10_N11 ds_grid_desc_m0_m10_m11_n0_n10_n11_;
+        EGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11_;
+
+        DefaultBlock2CTileMap block_2_ctile_map_;
+
+        // TODO: unused since gridwise_gemm_dl_v1r3 does NOT support prologue for the time being.
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceGemmMultipleD_Dl::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            {
+                std::cout << "arg.a_grid_desc_k0_m0_m1_k1_{"
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I0) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.b_grid_desc_k0_n0_n1_k1_{"
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I0) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.e_grid_desc_m_n_{ " << arg.e_grid_desc_m_n_.GetLength(I0) << ", "
+                          << arg.e_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+            }
+
+            if(!GridwiseGemm::CheckValidity(
+                   arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemmDlMultipleD_km_kn_mn has invalid setting");
+            }
+
+            const index_t grid_size = GridwiseGemm::CalculateGridSize(
+                arg.e_grid_desc_m_n_.GetLength(I0), arg.e_grid_desc_m_n_.GetLength(I1));
+
+            auto launch_kernel = [&](auto has_main_k_block_loop,
+                                     auto has_double_tail_k_block_loop) {
+                constexpr bool has_main_loop   = has_main_k_block_loop.value;
+                constexpr bool has_double_loop = has_double_tail_k_block_loop.value;
+
+                const auto kernel =
+                    kernel_gemm_dl_multiple_d<GridwiseGemm,
+                                              ADataType,
+                                              typename GridwiseGemm::DsGridPointer,
+                                              EDataType,
+                                              AElementwiseOperation,
+                                              BElementwiseOperation,
+                                              CDEElementwiseOperation,
+                                              DeviceOp::AGridDesc_K0_M0_M1_K1,
+                                              DeviceOp::BGridDesc_K0_N0_N1_K1,
+                                              DeviceOp::DsGridDesc_M0_M10_M11_N0_N10_N11,
+                                              DeviceOp::EGridDesc_M0_M10_M11_N0_N10_N11,
+                                              DefaultBlock2CTileMap,
+                                              has_main_loop,
+                                              has_double_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_a_grid_,
+                                              arg.p_b_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.a_grid_desc_k0_m0_m1_k1_,
+                                              arg.b_grid_desc_k0_n0_n1_k1_,
+                                              arg.ds_grid_desc_m0_m10_m11_n0_n10_n11_,
+                                              arg.e_grid_desc_m0_m10_m11_n0_n10_n11_,
+                                              arg.block_2_ctile_map_);
+            };
+
+            const auto K0                    = arg.a_grid_desc_k0_m0_m1_k1_.GetLength(I0);
+            const bool has_main_k_block_loop = GridwiseGemm::CalculateHasMainKBlockLoop(K0);
+            const bool has_double_tail_k_block_loop =
+                GridwiseGemm::CalculateHasDoubleTailKBlockLoop(K0);
+
+            if(has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, true>{});
+            }
+            else if(has_main_k_block_loop && !has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, false>{});
+            }
+            else if(!has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
+           ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030" ||
+           ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx1100" ||
+           ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102" ||
+           ck::get_device_name() == "gfx941" || ck::get_device_name() == "gfx942")
+        {
+            return GridwiseGemm::CheckValidity(
+                arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_);
+        }
+        else
+        {
+            return false;
+        }
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_a,
+                             const void* p_b,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t StrideA,
+                             index_t StrideB,
+                             std::array<ck::index_t, NumDTensor> StrideDs,
+                             index_t StrideE,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_e,
+                        M,
+                        N,
+                        K,
+                        StrideA,
+                        StrideB,
+                        StrideDs,
+                        StrideE,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        index_t M,
+                        index_t N,
+                        index_t K,
+                        index_t StrideA,
+                        index_t StrideB,
+                        std::array<ck::index_t, NumDTensor> StrideDs,
+                        index_t StrideE,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          M,
+                                          N,
+                                          K,
+                                          StrideA,
+                                          StrideB,
+                                          StrideDs,
+                                          StrideE,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGemmMultipleD_Dl"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << M1PerThread << ", "
+            << N1PerThread << ", "
+            << KPerThread
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..5afeaf98c42c74e4f88f13698bf7acbd904a1450
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -0,0 +1,1087 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemmWelford,
+          typename ABDataType,
+          typename DsPointer,
+          typename EMeanVarDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AGridDesc_AK0_M_AK1,
+          typename BGridDesc_BK0_N_BK1,
+          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename MeanVarGridDescriptor_MBlock_MPerBlock_NBlock,
+          typename CountGridDescriptor_MBlock_MPerBlock_NBlock,
+          typename Block2ETileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_multiple_d_welford_first_half_xdl_cshuffle(
+            const ABDataType* __restrict__ p_a_grid,
+            const ABDataType* __restrict__ p_b_grid,
+            DsPointer p_ds_grid,
+            EMeanVarDataType* __restrict__ p_e_grid,
+            EMeanVarDataType* __restrict__ p_welford_mean_grid,
+            EMeanVarDataType* __restrict__ p_welford_var_grid,
+            int32_t* __restrict__ p_welford_count_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1,
+            const BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1,
+            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+            const MeanVarGridDescriptor_MBlock_MPerBlock_NBlock
+                mean_var_grid_desc_mblock_mperblock_nblock,
+            const CountGridDescriptor_MBlock_MPerBlock_NBlock
+                count_grid_desc_mblock_mperblock_nblock,
+            const Block2ETileMap block_2_etile_map,
+            index_t NRaw)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemmWelford::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemmWelford::template Run<HasMainKBlockLoop>(
+        p_a_grid,
+        p_b_grid,
+        p_ds_grid,
+        p_e_grid,
+        p_welford_mean_grid,
+        p_welford_var_grid,
+        p_welford_count_grid,
+        p_shared,
+        a_element_op,
+        b_element_op,
+        cde_element_op,
+        a_grid_desc_ak0_m_ak1,
+        b_grid_desc_bk0_n_bk1,
+        ds_grid_desc_mblock_mperblock_nblock_nperblock,
+        e_grid_desc_mblock_mperblock_nblock_nperblock,
+        mean_var_grid_desc_mblock_mperblock_nblock,
+        count_grid_desc_mblock_mperblock_nblock,
+        block_2_etile_map,
+        NRaw);
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = p_welford_mean_grid;
+    ignore = p_welford_var_grid;
+    ignore = p_welford_count_grid;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = a_grid_desc_ak0_m_ak1;
+    ignore = b_grid_desc_bk0_n_bk1;
+    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = mean_var_grid_desc_mblock_mperblock_nblock;
+    ignore = count_grid_desc_mblock_mperblock_nblock;
+    ignore = block_2_etile_map;
+    ignore = NRaw;
+#endif
+}
+
+template <typename GridwiseWelfordLayernorm,
+          typename EMeanVarDataType,
+          typename HDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename ComputeDataType,
+          typename EHGridDesc_M_N,
+          typename LayernormMeanVarGridDesc_M_NBlock,
+          typename LayernormCountGridDesc_M_NBlock,
+          typename GammaBetaGridDesc_N,
+          typename HElementwiseOperation>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_welford_layernorm2d_second_half(
+            const EMeanVarDataType* __restrict__ p_e_grid,
+            const EMeanVarDataType* __restrict__ p_in_welford_mean_grid,
+            const EMeanVarDataType* __restrict__ p_in_welford_var_grid,
+            const int32_t* __restrict__ p_in_welford_count_grid,
+            const GammaDataType* __restrict__ p_gamma_grid,
+            const BetaDataType* __restrict__ p_beta_grid,
+            HDataType* __restrict__ p_h_grid,
+            const EHGridDesc_M_N e_grid_desc_m_n,
+            const EHGridDesc_M_N h_grid_desc_m_n,
+            const LayernormMeanVarGridDesc_M_NBlock mean_var_grid_desc_m_nblock,
+            const LayernormCountGridDesc_M_NBlock count_grid_desc_m_nblock,
+            const GammaBetaGridDesc_N gamma_grid_desc_n,
+            const GammaBetaGridDesc_N beta_grid_desc_n,
+            index_t numMeanVarCountBlockTileIteration_N,
+            index_t NBlockClusterLength,
+            ComputeDataType epsilon,
+            HElementwiseOperation h_element_op)
+{
+    GridwiseWelfordLayernorm::Run(p_e_grid,
+                                  p_in_welford_mean_grid,
+                                  p_in_welford_var_grid,
+                                  p_in_welford_count_grid,
+                                  p_gamma_grid,
+                                  p_beta_grid,
+                                  p_h_grid,
+                                  e_grid_desc_m_n,
+                                  h_grid_desc_m_n,
+                                  mean_var_grid_desc_m_nblock,
+                                  count_grid_desc_m_nblock,
+                                  gamma_grid_desc_n,
+                                  beta_grid_desc_n,
+                                  numMeanVarCountBlockTileIteration_N,
+                                  NBlockClusterLength,
+                                  epsilon,
+                                  h_element_op);
+}
+
+} // namespace ck
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   output : H[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+//   H = layernorm(E)
+// Assume:
+//   D0, D1, ... and E have the same layout
+//   Calculate mean & variance along N dimension in layernorm(E)
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename HLayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EMeanVarDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename HDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename HElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t GemmMPerBlock,
+          index_t GemmNPerBlock,
+          index_t GemmKPerBlock,
+          index_t AK1,
+          index_t BK1,
+          index_t MPerXDL,
+          index_t NPerXDL,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename PostShuffleThreadClusterSize_M_N,
+          index_t PostShuffleScalarPerVector,
+          typename LayernormThreadClusterSize_M_N,
+          index_t LayernormThreadSliceSize_M,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct DeviceGemmMultipleDLayernorm_Xdl_CShuffle
+    : public DeviceGemmMultipleDLayernorm<ALayout,
+                                          BLayout,
+                                          DsLayout,
+                                          HLayout,
+                                          ADataType,
+                                          BDataType,
+                                          DsDataType,
+                                          GammaDataType,
+                                          BetaDataType,
+                                          HDataType,
+                                          AElementwiseOperation,
+                                          BElementwiseOperation,
+                                          CDEElementwiseOperation,
+                                          HElementwiseOperation>
+{
+    // EDataType, MeanDataType and VarDataType must be the same.
+    // eg. M, N, K = [1, 1, 1],
+    // in case of layernorm, divisor = 1 / sqrt(var + 1e-5) = 316.227783
+    // if (x - mean) != 0, (x - mean) * divisor * gamma might be too large
+    // However, (x - mean) * divisor * gamma should be 0 in this case
+
+    using DeviceOp = DeviceGemmMultipleDLayernorm_Xdl_CShuffle;
+    using ELayout  = HLayout;
+
+    static constexpr index_t NumDTensor                  = DsDataType::Size();
+    static constexpr index_t LayernormHDstVectorSize     = PostShuffleScalarPerVector;
+    static constexpr index_t LayernormGammaSrcVectorSize = PostShuffleScalarPerVector;
+    static constexpr index_t LayernormBetaSrcVectorSize  = PostShuffleScalarPerVector;
+    static constexpr index_t LayernormESrcVectorSize     = PostShuffleScalarPerVector;
+    static constexpr index_t LayernormThreadSliceSize_N  = PostShuffleScalarPerVector;
+    using LayernormBlockTileSize_M_N =
+        Sequence<LayernormThreadClusterSize_M_N::At(0) * LayernormThreadSliceSize_M,
+                 LayernormThreadClusterSize_M_N::At(1) * LayernormThreadSliceSize_N>;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+
+    static constexpr auto matrix_padder = MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+        GemmMPerBlock, GemmNPerBlock, GemmKPerBlock};
+
+    static auto MakeAGridDescriptor_M_K(index_t MRaw, index_t KRaw, index_t StrideA)
+    {
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(I1, StrideA));
+            }
+        }();
+
+        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+    }
+
+    static auto MakeBGridDescriptor_N_K(index_t KRaw, index_t NRaw, index_t StrideB)
+    {
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(StrideB, I1));
+            }
+        }();
+
+        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+    }
+
+    template <typename DoPads, index_t MPerTile, index_t NPerTile>
+    static auto MakeEHGridDescriptor_M_N(index_t M, index_t N, index_t Stride)
+    {
+        // Only support row major for E and H
+        const auto grid_desc_m_n =
+            make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(Stride, I1));
+        return PadTensorDescriptor(grid_desc_m_n, make_tuple(MPerTile, NPerTile), DoPads{});
+    }
+
+    static auto MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                                         const std::array<index_t, NumDTensor>& NRaws,
+                                         const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                static_assert(is_same<tensor_layout::gemm::RowMajor, DLayout>::value);
+
+                return DeviceOp::
+                    MakeEHGridDescriptor_M_N<Sequence<true, true>, GemmMPerBlock, GemmNPerBlock>(
+                        MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    template <typename DoPads, index_t MPerTile, index_t NPerTile>
+    static auto MakeMeanVarDescriptor_M_N(index_t M, index_t N)
+    {
+        const auto grid_desc_m_n =
+            make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(N, I1));
+        return PadTensorDescriptor(grid_desc_m_n, make_tuple(MPerTile, NPerTile), DoPads{});
+    }
+
+    template <typename DoPads, index_t MPerTile, index_t NPerTile>
+    static auto MakeCountDescriptor_M_N(index_t M, index_t N)
+    {
+        // We will broadcast [N] to [M, N] in this descriptor
+        // Hence, 1st stride is 0
+        const auto grid_desc_m_n =
+            make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I0, I1));
+        return PadTensorDescriptor(grid_desc_m_n, make_tuple(MPerTile, NPerTile), DoPads{});
+    }
+
+    template <index_t XPerTile>
+    static auto MakeDescriptor_X(index_t X)
+    {
+        const auto grid_desc_x = make_naive_tensor_descriptor_packed(make_tuple(X));
+        return PadTensorDescriptor(grid_desc_x, make_tuple(XPerTile), Sequence<true>{});
+    }
+
+    using AGridDesc_M_K  = decltype(MakeAGridDescriptor_M_K(1, 1, 1));
+    using BGridDesc_N_K  = decltype(MakeBGridDescriptor_N_K(1, 1, 1));
+    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
+    // We have to separate mean var descriptor for gemm and layernorm bacause of different grid
+    // layout(different padding)
+    using GemmMeanVarGridDesc_M_NBlock = decltype(
+        MakeMeanVarDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1, 1));
+
+    using GemmCountGridDesc_M_NBlock = decltype(
+        MakeCountDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1, 1));
+
+    using LayernormMeanVarGridDesc_M_NBlock =
+        decltype(MakeMeanVarDescriptor_M_N<Sequence<true, true>,
+                                           LayernormBlockTileSize_M_N::At(0),
+                                           LayernormBlockTileSize_M_N::At(1)>(1, 1));
+
+    using LayernormCountGridDesc_M_NBlock =
+        decltype(MakeCountDescriptor_M_N<Sequence<true, true>,
+                                         LayernormBlockTileSize_M_N::At(0),
+                                         LayernormBlockTileSize_M_N::At(1)>(1, 1));
+
+    using GammaBetaGridDesc_N = decltype(MakeDescriptor_X<LayernormBlockTileSize_M_N::At(1)>(1));
+    using EHGridDesc_M_N = decltype(MakeEHGridDescriptor_M_N<Sequence<true, true>, 1, 1>(1, 1, 1));
+
+    using GridwiseGemmWelford = GridwiseGemmMultipleDWelfordFirstHalf_xdl_cshuffle<
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EMeanVarDataType,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        AGridDesc_M_K,
+        BGridDesc_N_K,
+        DsGridDesc_M_N,
+        EHGridDesc_M_N,
+        GemmMeanVarGridDesc_M_NBlock,
+        GemmCountGridDesc_M_NBlock,
+        NumGemmKPrefetchStage,
+        BlockSize,
+        GemmMPerBlock,
+        GemmNPerBlock,
+        GemmKPerBlock,
+        AK1,
+        BK1,
+        MPerXDL,
+        NPerXDL,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        PostShuffleThreadClusterSize_M_N,
+        PostShuffleScalarPerVector,
+        LoopSched,
+        PipelineVer>;
+
+    using Block2ETileMap = typename GridwiseGemmWelford::DefaultBlock2ETileMap;
+
+    using GridwiseWelfordLayernorm =
+        GridwiseWelfordSecondHalfLayernorm2d<EMeanVarDataType,
+                                             HDataType,
+                                             GammaDataType,
+                                             BetaDataType,
+                                             AccDataType,
+                                             EHGridDesc_M_N,
+                                             LayernormMeanVarGridDesc_M_NBlock,
+                                             LayernormCountGridDesc_M_NBlock,
+                                             GammaBetaGridDesc_N,
+                                             HElementwiseOperation,
+                                             BlockSize,
+                                             LayernormThreadClusterSize_M_N::At(I0),
+                                             LayernormThreadClusterSize_M_N::At(I1),
+                                             LayernormThreadSliceSize_M,
+                                             LayernormThreadSliceSize_N,
+                                             LayernormESrcVectorSize,
+                                             LayernormHDstVectorSize,
+                                             LayernormGammaSrcVectorSize,
+                                             LayernormBetaSrcVectorSize>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a_grid,
+                 const void* p_b_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 const void* p_gamma_grid,
+                 const void* p_beta_grid,
+                 void* p_h_grid,
+                 index_t MRaw,
+                 index_t NRaw,
+                 index_t KRaw,
+                 index_t StrideA,
+                 index_t StrideB,
+                 std::array<index_t, NumDTensor> StrideDs,
+                 index_t StrideH,
+                 double epsilon,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op,
+                 HElementwiseOperation h_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
+              p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
+              p_ds_grid_{},
+              p_workspace_e_grid_{nullptr},
+              p_workspace_mean_{nullptr},
+              p_workspace_var_{nullptr},
+              p_workspace_count_{nullptr},
+              p_gamma_grid_{static_cast<const GammaDataType*>(p_gamma_grid)},
+              p_beta_grid_{static_cast<const BetaDataType*>(p_beta_grid)},
+              p_h_grid_{static_cast<HDataType*>(p_h_grid)},
+              a_grid_desc_m_k_{DeviceOp::MakeAGridDescriptor_M_K(MRaw, KRaw, StrideA)},
+              b_grid_desc_n_k_{DeviceOp::MakeBGridDescriptor_N_K(KRaw, NRaw, StrideB)},
+              ds_grid_desc_m_n_{},
+              gemm_e_grid_desc_m_n_{
+                  DeviceOp::MakeEHGridDescriptor_M_N<Sequence<true, true>,
+                                                     GemmMPerBlock,
+                                                     GemmNPerBlock>(MRaw, NRaw, StrideH)},
+              layernorm_e_grid_desc_m_n_{
+                  DeviceOp::MakeEHGridDescriptor_M_N<Sequence<true, true>,
+                                                     LayernormBlockTileSize_M_N::At(0),
+                                                     LayernormBlockTileSize_M_N::At(1)>(
+                      MRaw, NRaw, StrideH)},
+              gemm_mean_var_grid_desc_m_nblock_{},
+              gemm_count_grid_desc_m_nblock_{},
+              layernorm_mean_var_grid_desc_m_nblock_{},
+              layernorm_count_grid_desc_m_nblock_{},
+              gamma_grid_desc_n_{
+                  DeviceOp::MakeDescriptor_X<LayernormBlockTileSize_M_N::At(1)>(NRaw)},
+              beta_grid_desc_n_{
+                  DeviceOp::MakeDescriptor_X<LayernormBlockTileSize_M_N::At(1)>(NRaw)},
+              h_grid_desc_m_n_{
+                  DeviceOp::MakeEHGridDescriptor_M_N<Sequence<true, true>,
+                                                     LayernormBlockTileSize_M_N::At(0),
+                                                     LayernormBlockTileSize_M_N::At(1)>(
+                      MRaw, NRaw, StrideH)},
+              a_grid_desc_ak0_m_ak1_{
+                  GridwiseGemmWelford::MakeDefaultAGridDescriptor_AK0_M_AK1(a_grid_desc_m_k_)},
+              b_grid_desc_bk0_n_bk1_{
+                  GridwiseGemmWelford::MakeDefaultBGridDescriptor_BK0_N_BK1(b_grid_desc_n_k_)},
+              block_2_etile_map_{
+                  GridwiseGemmWelford::MakeDefaultBlock2ETileMap(gemm_e_grid_desc_m_n_)},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op},
+              h_element_op_{h_element_op},
+              MRaw_{MRaw},
+              NRaw_{NRaw},
+              KRaw_{KRaw},
+              gemm_nblock_{math::integer_divide_ceil(NRaw, GemmNPerBlock)},
+              epsilon_{static_cast<AccDataType>(epsilon)}
+        {
+            // We don't need to pad in N dimension in gemm for mean/var/count. Set NPerTile 1.
+            gemm_mean_var_grid_desc_m_nblock_ =
+                DeviceOp::MakeMeanVarDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, 1>(
+                    MRaw, gemm_nblock_);
+
+            gemm_count_grid_desc_m_nblock_ =
+                DeviceOp::MakeCountDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, 1>(
+                    MRaw, gemm_nblock_);
+
+            layernorm_mean_var_grid_desc_m_nblock_ =
+                DeviceOp::MakeMeanVarDescriptor_M_N<Sequence<true, true>,
+                                                    LayernormBlockTileSize_M_N::At(0),
+                                                    LayernormBlockTileSize_M_N::At(1)>(
+                    MRaw, gemm_nblock_);
+
+            layernorm_count_grid_desc_m_nblock_ =
+                DeviceOp::MakeCountDescriptor_M_N<Sequence<true, true>,
+                                                  LayernormBlockTileSize_M_N::At(0),
+                                                  LayernormBlockTileSize_M_N::At(1)>(MRaw,
+                                                                                     gemm_nblock_);
+
+            // populate pointer, desc for Ds
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    DeviceOp::MakeEHGridDescriptor_M_N<Sequence<true, true>,
+                                                       GemmMPerBlock,
+                                                       GemmNPerBlock>(MRaw, NRaw, StrideDs[i]);
+            });
+
+            // populate desc for Ds/E/mean/var/count
+            if(GridwiseGemmWelford::CheckValidity(a_grid_desc_m_k_,
+                                                  b_grid_desc_n_k_,
+                                                  ds_grid_desc_m_n_,
+                                                  gemm_e_grid_desc_m_n_,
+                                                  block_2_etile_map_))
+            {
+                ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemmWelford::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        ds_grid_desc_m_n_);
+
+                e_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemmWelford::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        gemm_e_grid_desc_m_n_);
+
+                gemm_mean_var_grid_desc_mblock_mperblock_nblock_ =
+                    GridwiseGemmWelford::MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(
+                        gemm_mean_var_grid_desc_m_nblock_);
+
+                gemm_count_grid_desc_mblock_mperblock_nblock_ =
+                    GridwiseGemmWelford::MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(
+                        gemm_count_grid_desc_m_nblock_);
+            }
+        }
+
+        void Print() const
+        {
+            std::cout << "A[M, K]: " << a_grid_desc_m_k_ << std::endl;
+            std::cout << "B[N, K]: " << b_grid_desc_n_k_ << std::endl;
+            static_for<0, NumDTensor, 1>{}(
+                [&](auto i) { std::cout << "Ds[M, N]: " << ds_grid_desc_m_n_[i] << std::endl; });
+            std::cout << "E[M, N]: " << gemm_e_grid_desc_m_n_ << std::endl;
+            std::cout << "H[M, N]: " << h_grid_desc_m_n_ << std::endl;
+        }
+
+        //  private:
+        // pointers
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseGemmWelford::DsGridPointer p_ds_grid_;
+        void* p_workspace_e_grid_;
+        void* p_workspace_mean_;
+        void* p_workspace_var_;
+        void* p_workspace_count_;
+        const GammaDataType* p_gamma_grid_;
+        const BetaDataType* p_beta_grid_;
+        HDataType* p_h_grid_;
+
+        // tensor descriptors for problem definiton
+        AGridDesc_M_K a_grid_desc_m_k_;
+        BGridDesc_N_K b_grid_desc_n_k_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EHGridDesc_M_N gemm_e_grid_desc_m_n_;
+        EHGridDesc_M_N layernorm_e_grid_desc_m_n_;
+        GemmMeanVarGridDesc_M_NBlock gemm_mean_var_grid_desc_m_nblock_;
+        GemmCountGridDesc_M_NBlock gemm_count_grid_desc_m_nblock_;
+        LayernormMeanVarGridDesc_M_NBlock layernorm_mean_var_grid_desc_m_nblock_;
+        LayernormCountGridDesc_M_NBlock layernorm_count_grid_desc_m_nblock_;
+        GammaBetaGridDesc_N gamma_grid_desc_n_;
+        GammaBetaGridDesc_N beta_grid_desc_n_;
+        EHGridDesc_M_N h_grid_desc_m_n_;
+
+        // tensor descriptors for block/thread-wise copy
+        typename GridwiseGemmWelford::DefaultAGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
+        typename GridwiseGemmWelford::DefaultBGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
+        typename GridwiseGemmWelford::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
+        typename GridwiseGemmWelford::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            e_grid_desc_mblock_mperblock_nblock_nperblock_;
+        typename GridwiseGemmWelford::MeanVarGridDescriptor_MBlock_MPerBlock_NBlock
+            gemm_mean_var_grid_desc_mblock_mperblock_nblock_;
+        typename GridwiseGemmWelford::CountGridDescriptor_MBlock_MPerBlock_NBlock
+            gemm_count_grid_desc_mblock_mperblock_nblock_;
+
+        // block-to-e-tile map
+        Block2ETileMap block_2_etile_map_;
+
+        // element-wise op
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+        HElementwiseOperation h_element_op_;
+
+        index_t MRaw_;
+        index_t NRaw_;
+        index_t KRaw_;
+        index_t gemm_nblock_;
+        AccDataType epsilon_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            float avg_time = 0;
+
+            if(!GridwiseGemmWelford::CheckValidity(arg.a_grid_desc_m_k_,
+                                                   arg.b_grid_desc_n_k_,
+                                                   arg.ds_grid_desc_m_n_,
+                                                   arg.gemm_e_grid_desc_m_n_,
+                                                   arg.block_2_etile_map_))
+            {
+                throw std::runtime_error("wrong! GridwiseGemmWelford has invalid setting");
+            }
+            if(arg.p_workspace_e_grid_ == nullptr || arg.p_workspace_mean_ == nullptr ||
+               arg.p_workspace_var_ == nullptr || arg.p_workspace_count_ == nullptr)
+                throw std::runtime_error("wrong! WorkSpace pointer has not been set");
+
+            index_t grid_size = arg.block_2_etile_map_.CalculateGridSize(arg.gemm_e_grid_desc_m_n_);
+
+            const auto M = arg.h_grid_desc_m_n_.GetLength(I0);
+            const auto N = arg.h_grid_desc_m_n_.GetLength(I1);
+            const auto K =
+                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel_gemm_welford =
+                    kernel_gemm_multiple_d_welford_first_half_xdl_cshuffle<
+                        GridwiseGemmWelford,
+                        ADataType, // TODO: distiguish A/B datatype
+                        typename GridwiseGemmWelford::DsGridPointer,
+                        EMeanVarDataType,
+                        AElementwiseOperation,
+                        BElementwiseOperation,
+                        CDEElementwiseOperation,
+                        typename GridwiseGemmWelford::DefaultAGridDesc_AK0_M_AK1,
+                        typename GridwiseGemmWelford::DefaultBGridDesc_BK0_N_BK1,
+                        typename GridwiseGemmWelford::
+                            DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                        typename GridwiseGemmWelford::
+                            EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                        typename GridwiseGemmWelford::MeanVarGridDescriptor_MBlock_MPerBlock_NBlock,
+                        typename GridwiseGemmWelford::CountGridDescriptor_MBlock_MPerBlock_NBlock,
+                        typename GridwiseGemmWelford::DefaultBlock2ETileMap,
+                        has_main_loop>;
+
+                const auto kernel_welford_layernorm =
+                    kernel_welford_layernorm2d_second_half<GridwiseWelfordLayernorm,
+                                                           EMeanVarDataType,
+                                                           HDataType,
+                                                           GammaDataType,
+                                                           BetaDataType,
+                                                           AccDataType,
+                                                           EHGridDesc_M_N,
+                                                           LayernormMeanVarGridDesc_M_NBlock,
+                                                           LayernormCountGridDesc_M_NBlock,
+                                                           GammaBetaGridDesc_N,
+                                                           HElementwiseOperation>;
+
+                avg_time +=
+                    launch_and_time_kernel(stream_config,
+                                           kernel_gemm_welford,
+                                           dim3(grid_size),
+                                           dim3(BlockSize),
+                                           0,
+                                           arg.p_a_grid_,
+                                           arg.p_b_grid_,
+                                           arg.p_ds_grid_,
+                                           static_cast<EMeanVarDataType*>(arg.p_workspace_e_grid_),
+                                           static_cast<EMeanVarDataType*>(arg.p_workspace_mean_),
+                                           static_cast<EMeanVarDataType*>(arg.p_workspace_var_),
+                                           static_cast<int32_t*>(arg.p_workspace_count_),
+                                           arg.a_element_op_,
+                                           arg.b_element_op_,
+                                           arg.cde_element_op_,
+                                           arg.a_grid_desc_ak0_m_ak1_,
+                                           arg.b_grid_desc_bk0_n_bk1_,
+                                           arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                           arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                           arg.gemm_mean_var_grid_desc_mblock_mperblock_nblock_,
+                                           arg.gemm_count_grid_desc_mblock_mperblock_nblock_,
+                                           arg.block_2_etile_map_,
+                                           arg.NRaw_);
+
+                index_t MBlockClusterLength =
+                    math::integer_divide_ceil(M, LayernormBlockTileSize_M_N::At(0));
+                index_t NBlockClusterLength =
+                    math::integer_divide_ceil(N, LayernormBlockTileSize_M_N::At(1));
+                grid_size = MBlockClusterLength * NBlockClusterLength;
+
+                index_t numMeanVarCountBlockTileIteration_N = math::integer_divide_ceil(
+                    arg.gemm_nblock_, LayernormThreadClusterSize_M_N::At(I1));
+
+                avg_time += launch_and_time_kernel(
+                    stream_config,
+                    kernel_welford_layernorm,
+                    dim3(grid_size),
+                    dim3(BlockSize),
+                    0,
+                    static_cast<EMeanVarDataType*>(arg.p_workspace_e_grid_),
+                    static_cast<const EMeanVarDataType*>(arg.p_workspace_mean_),
+                    static_cast<const EMeanVarDataType*>(arg.p_workspace_var_),
+                    static_cast<const int32_t*>(arg.p_workspace_count_),
+                    arg.p_gamma_grid_,
+                    arg.p_beta_grid_,
+                    arg.p_h_grid_,
+                    arg.layernorm_e_grid_desc_m_n_,
+                    arg.h_grid_desc_m_n_,
+                    arg.layernorm_mean_var_grid_desc_m_nblock_,
+                    arg.layernorm_count_grid_desc_m_nblock_,
+                    arg.gamma_grid_desc_n_,
+                    arg.beta_grid_desc_n_,
+                    numMeanVarCountBlockTileIteration_N,
+                    NBlockClusterLength,
+                    arg.epsilon_,
+                    arg.h_element_op_);
+
+                return avg_time;
+            };
+
+            if(GridwiseGemmWelford::CalculateHasMainKBlockLoop(K))
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        size_t workspace_size = 0;
+
+        int gemm_welford_size = pArg_->MRaw_ * pArg_->gemm_nblock_;
+
+        // workspace for welford intermediate mean
+        workspace_size += gemm_welford_size * sizeof(EMeanVarDataType) + 64;
+
+        // workspace for welford intermediate variance
+        workspace_size += gemm_welford_size * sizeof(EMeanVarDataType) + 64;
+
+        // workspace for welford intermediate count
+        workspace_size += pArg_->gemm_nblock_ * sizeof(int32_t) + 64;
+
+        if constexpr(!is_same_v<EMeanVarDataType, HDataType>)
+            workspace_size += pArg_->MRaw_ * pArg_->NRaw_ * sizeof(EMeanVarDataType);
+
+        return (workspace_size);
+    };
+
+    void SetWorkSpacePointer(BaseArgument* pArg, void* p_workspace) const override
+    {
+        Argument* pArg_ = dynamic_cast<Argument*>(pArg);
+
+        pArg_->p_workspace_ = p_workspace;
+
+        int gemm_welford_size = pArg_->MRaw_ * pArg_->gemm_nblock_;
+
+        // setup buffer used for intermediate welford mean
+        pArg_->p_workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
+
+        index_t mean_space_sz = gemm_welford_size * sizeof(EMeanVarDataType);
+        mean_space_sz         = math::integer_least_multiple(mean_space_sz, 64);
+
+        // setup buffer used for intermediate welford varirance
+        pArg_->p_workspace_var_ = reinterpret_cast<char*>(pArg_->p_workspace_mean_) + mean_space_sz;
+
+        index_t variance_space_sz = gemm_welford_size * sizeof(EMeanVarDataType);
+        variance_space_sz         = math::integer_least_multiple(variance_space_sz, 64);
+
+        // setup buffer used for intermediate welford count
+        pArg_->p_workspace_count_ =
+            reinterpret_cast<char*>(pArg_->p_workspace_var_) + variance_space_sz;
+
+        index_t count_space_sz = gemm_welford_size * sizeof(int32_t);
+        count_space_sz         = math::integer_least_multiple(count_space_sz, 64);
+
+        if constexpr(!is_same_v<EMeanVarDataType, HDataType>)
+            pArg_->p_workspace_e_grid_ =
+                reinterpret_cast<char*>(pArg_->p_workspace_count_) + count_space_sz;
+        else
+            pArg_->p_workspace_e_grid_ = static_cast<void*>(pArg_->p_h_grid_);
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
+        {
+            return false;
+        }
+
+        // check vector load/store
+        {
+            using Row = ck::tensor_layout::gemm::RowMajor;
+            using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+            // check vector load of A
+            if constexpr(is_same_v<ALayout, Row> && ABlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Col> && ABlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.MRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector laod of B
+            if constexpr(is_same_v<BLayout, Col> && BBlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<BLayout, Row> && BBlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.NRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector load of Ds
+            // only support RowMajor for now
+            bool all_valid = true;
+
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                if constexpr(!is_same_v<DLayout, Row>)
+                {
+                    all_valid = false;
+                }
+            });
+
+            if(!all_valid)
+            {
+                return false;
+            }
+
+            // check vector store of E
+            // E and H only support RowMajor for now
+            if constexpr(is_same_v<ELayout, Row> && is_same_v<HLayout, Row>)
+            {
+                if(arg.NRaw_ % PostShuffleScalarPerVector != 0 ||
+                   arg.NRaw_ % LayernormGammaSrcVectorSize != 0 ||
+                   arg.NRaw_ % LayernormBetaSrcVectorSize != 0 ||
+                   arg.NRaw_ % LayernormHDstVectorSize != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+        }
+
+        return GridwiseGemmWelford::CheckValidity(arg.a_grid_desc_m_k_,
+                                                  arg.b_grid_desc_n_k_,
+                                                  arg.ds_grid_desc_m_n_,
+                                                  arg.gemm_e_grid_desc_m_n_,
+                                                  arg.block_2_etile_map_);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_a,
+                             const void* p_b,
+                             std::array<const void*, NumDTensor> p_ds,
+                             const void* p_gamma,
+                             const void* p_beta,
+                             void* p_h,
+                             index_t MRaw,
+                             index_t NRaw,
+                             index_t KRaw,
+                             index_t StrideA,
+                             index_t StrideB,
+                             std::array<index_t, NumDTensor> StrideDs,
+                             index_t StrideH,
+                             double epsilon,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op,
+                             HElementwiseOperation h_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_gamma,
+                        p_beta,
+                        p_h,
+                        MRaw,
+                        NRaw,
+                        KRaw,
+                        StrideA,
+                        StrideB,
+                        StrideDs,
+                        StrideH,
+                        epsilon,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op,
+                        h_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
+                                                      const void* p_b,
+                                                      std::array<const void*, NumDTensor> p_ds,
+                                                      const void* p_gamma,
+                                                      const void* p_beta,
+                                                      void* p_h,
+                                                      index_t MRaw,
+                                                      index_t NRaw,
+                                                      index_t KRaw,
+                                                      index_t StrideA,
+                                                      index_t StrideB,
+                                                      std::array<index_t, NumDTensor> StrideDs,
+                                                      index_t StrideH,
+                                                      double epsilon,
+                                                      AElementwiseOperation a_element_op,
+                                                      BElementwiseOperation b_element_op,
+                                                      CDEElementwiseOperation cde_element_op,
+                                                      HElementwiseOperation h_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_gamma,
+                                          p_beta,
+                                          p_h,
+                                          MRaw,
+                                          NRaw,
+                                          KRaw,
+                                          StrideA,
+                                          StrideB,
+                                          StrideDs,
+                                          StrideH,
+                                          epsilon,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op,
+                                          h_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<LoopScheduler, std::string> LoopSchedToString{
+            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceGemmMultipleDLayernorm_Xdl_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << GemmMPerBlock << ", "
+            << GemmNPerBlock << ", "
+            << GemmKPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << getGemmSpecializationString(GemmSpec) << ", "
+            << PostShuffleThreadClusterSize_M_N::At(I0) << ", "
+            << PostShuffleThreadClusterSize_M_N::At(I1) << ", "
+            << LayernormThreadClusterSize_M_N::At(I0) << ", "
+            << LayernormThreadClusterSize_M_N::At(I1) << ", "
+            << LayernormThreadSliceSize_M
+            << ">"
+            << " LoopScheduler: "
+            << LoopSchedToString[LoopSched] << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+}; // namespace device
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index f1fb4ab4b1b255e99c4c5a74c07f97019cf55d87..3a0be837b7f3aa6683b7ade07a3d32f71f1cc0b5 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -60,7 +60,8 @@ __global__ void
             const RsGridDescriptor_MBlock_MPerBlock rs_grid_desc_mblock_mperblock,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -554,7 +555,9 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
@@ -669,7 +672,15 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
             << KPerBlock << ", "
             << AK1 << ", "
             << BK1 << ", "
-            << getGemmSpecializationString(GemmSpec)
+            << getGemmSpecializationString(GemmSpec) << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..44b3518e2c950e15ec3d5cfd240c443029987622
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
@@ -0,0 +1,706 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t K0PerBlock,
+          ck::index_t K1,
+          ck::index_t MPerWMMA,
+          ck::index_t NPerWMMA,
+          ck::index_t MRepeat,
+          ck::index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsAddExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsAddExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+          ck::index_t NumPrefetch         = 1,
+          ck::LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
+struct DeviceGemmMultipleD_Wmma_CShuffle : public DeviceGemmMultipleD<ALayout,
+                                                                      BLayout,
+                                                                      DsLayout,
+                                                                      ELayout,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      DsDataType,
+                                                                      EDataType,
+                                                                      AElementwiseOperation,
+                                                                      BElementwiseOperation,
+                                                                      CDEElementwiseOperation>
+{
+    using DeviceOp                      = DeviceGemmMultipleD_Wmma_CShuffle;
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    // K1 = Max Vector Access Pixels
+    static constexpr auto K1Number = Number<K1>{};
+
+    static constexpr auto matrix_padder =
+        MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, K0PerBlock* K1};
+
+    static auto MakeAGridDescriptor_K0_M_K1(index_t MRaw, index_t KRaw, index_t StrideA)
+    {
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto a_grid_desc_m_k = matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+        const auto M               = a_grid_desc_m_k.GetLength(I0);
+        const auto K               = a_grid_desc_m_k.GetLength(I1);
+        assert(K % K1 == 0);
+        const index_t K0 = K / K1;
+
+        return transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                       make_pass_through_transform(M)),
+            make_tuple(Sequence<1>{}, Sequence<0>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    static auto MakeBGridDescriptor_K0_N_K1(index_t KRaw, index_t NRaw, index_t StrideB)
+    {
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(StrideB, I1));
+            }
+        }();
+
+        const auto b_grid_desc_n_k = matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+        const auto N               = b_grid_desc_n_k.GetLength(I0);
+        const auto K               = b_grid_desc_n_k.GetLength(I1);
+        assert(K % K1 == 0);
+        const index_t K0 = K / K1;
+
+        return transform_tensor_descriptor(
+            b_grid_desc_n_k,
+            make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                       make_pass_through_transform(N)),
+            make_tuple(Sequence<1>{}, Sequence<0>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    template <typename ELayout_>
+    static auto MakeEGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideE)
+    {
+        const auto e_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELayout_>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELayout_>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideE));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    static auto MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& Ms,
+                                         const std::array<index_t, NumDTensor>& Ns,
+                                         const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return DeviceOp::MakeEGridDescriptor_M_N<DLayout>(Ms[i], Ns[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // Gridwise descriptor, mapping to whole given provblem.
+    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
+    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
+    using DsGridDesc_M_N    = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
+    using EGridDesc_M_N     = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
+
+    // GridwiseOp
+    using GridwiseOp = GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle<
+        // DataType Family
+        ADataType,
+        BDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        // InMemory Data Descriptor
+        AGridDesc_K0_M_K1,
+        BGridDesc_K0_N_K1,
+        DsGridDesc_M_N,
+        EGridDesc_M_N,
+        // ElementwiseOp Family
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        // Tiling Family
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerWMMA,
+        NPerWMMA,
+        K1,
+        MRepeat,
+        NRepeat,
+        // ThreadCluster Family
+        BlockSize,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+        NumPrefetch,
+        LoopSched,
+        PipelineVer>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a_grid,
+                 const void* p_b_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 index_t M,
+                 index_t N,
+                 index_t K,
+                 index_t StrideA,
+                 index_t StrideB,
+                 std::array<index_t, NumDTensor> StrideDs,
+                 index_t StrideE,
+                 index_t M01,
+                 index_t N01,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
+              p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              a_grid_desc_k0_m_k1_{},
+              b_grid_desc_k0_n_k1_{},
+              ds_grid_desc_m_n_{},
+              e_grid_desc_m_n_{},
+              ds_grid_desc_mblock_mperblock_nblock_nperblock{},
+              e_grid_desc_mblock_mperblock_nblock_nperblock{},
+              block_2_ctile_map_{},
+              M01_{M01},
+              N01_{N01},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op},
+              MRaw_{M},
+              NRaw_{N},
+              KRaw_{K}
+        {
+            a_grid_desc_k0_m_k1_ = DeviceOp::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
+            b_grid_desc_k0_n_k1_ = DeviceOp::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    DeviceOp::MakeEGridDescriptor_M_N<DLayout>(M, N, StrideDs[i]);
+            });
+            e_grid_desc_m_n_ = DeviceOp::MakeEGridDescriptor_M_N<ELayout>(M, N, StrideE);
+
+            block_2_ctile_map_ = GridwiseOp::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_, M01, N01);
+
+            if(GridwiseOp::CheckValidity(a_grid_desc_k0_m_k1_,
+                                         b_grid_desc_k0_n_k1_,
+                                         ds_grid_desc_m_n_,
+                                         e_grid_desc_m_n_,
+                                         block_2_ctile_map_))
+            {
+                ds_grid_desc_mblock_mperblock_nblock_nperblock =
+                    GridwiseOp::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        ds_grid_desc_m_n_);
+
+                e_grid_desc_mblock_mperblock_nblock_nperblock =
+                    GridwiseOp::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        e_grid_desc_m_n_);
+            }
+        }
+
+        // Pointers
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseOp::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        // Tensor Descriptors
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+        typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            ds_grid_desc_mblock_mperblock_nblock_nperblock;
+        typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            e_grid_desc_mblock_mperblock_nblock_nperblock;
+
+        // Block to Tile mapping
+        typename GridwiseOp::DefaultBlock2CTileMap block_2_ctile_map_;
+
+        // Idle
+        index_t M01_;
+        index_t N01_;
+
+        // ElementwiseOp
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+
+        // for checking vector load/store
+        index_t MRaw_;
+        index_t NRaw_;
+        index_t KRaw_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+#if 0
+            {
+                std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
+                          << ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
+                          << ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) 
+                          << ", " << arg.c_grid_desc_m_n_.GetLength(I1) << ", "
+                          << arg.c_grid_desc_m_n_.GetLength(I2) << "}" << std::endl;
+            }
+#endif
+
+            if(!GridwiseOp::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                          arg.b_grid_desc_k0_n_k1_,
+                                          arg.ds_grid_desc_m_n_,
+                                          arg.e_grid_desc_m_n_,
+                                          arg.block_2_ctile_map_))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemm_k0mk1_k0nk1_m0nm1_wmma_v1r1 has invalid setting");
+            }
+
+            const index_t grid_size =
+                arg.block_2_ctile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
+
+            const auto K =
+                arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
+
+            float ave_time = 0;
+
+            if(GridwiseOp::CalculateHasMainKBlockLoop(K))
+            {
+                const auto kernel = kernel_gemm_mupltipe_d_wmma_cshuffle<
+                    GridwiseOp,
+                    ADataType,
+                    BDataType,
+                    typename GridwiseOp::DsGridPointer,
+                    EDataType,
+                    remove_reference_t<typename DeviceOp::AGridDesc_K0_M_K1>,
+                    remove_reference_t<typename DeviceOp::BGridDesc_K0_N_K1>,
+                    remove_reference_t<
+                        typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    remove_reference_t<
+                        typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CDEElementwiseOperation,
+                    remove_reference_t<typename GridwiseOp::DefaultBlock2CTileMap>,
+                    true>; // Last Option is W/O
+
+                ave_time =
+                    launch_and_time_kernel(stream_config,
+                                           kernel,
+                                           dim3(grid_size),
+                                           dim3(BlockSize),
+                                           0,
+                                           arg.p_a_grid_,
+                                           arg.p_b_grid_,
+                                           arg.p_ds_grid_,
+                                           arg.p_e_grid_,
+                                           arg.a_grid_desc_k0_m_k1_,
+                                           arg.b_grid_desc_k0_n_k1_,
+                                           arg.ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                           arg.e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                           arg.a_element_op_,
+                                           arg.b_element_op_,
+                                           arg.cde_element_op_,
+                                           arg.block_2_ctile_map_);
+            }
+            else
+            {
+                const auto kernel = kernel_gemm_mupltipe_d_wmma_cshuffle<
+                    GridwiseOp,
+                    ADataType,
+                    BDataType,
+                    typename GridwiseOp::DsGridPointer,
+                    EDataType,
+                    remove_reference_t<typename DeviceOp::AGridDesc_K0_M_K1>,
+                    remove_reference_t<typename DeviceOp::BGridDesc_K0_N_K1>,
+                    remove_reference_t<
+                        typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    remove_reference_t<
+                        typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CDEElementwiseOperation,
+                    remove_reference_t<typename GridwiseOp::DefaultBlock2CTileMap>,
+                    false>;
+
+                ave_time =
+                    launch_and_time_kernel(stream_config,
+                                           kernel,
+                                           dim3(grid_size),
+                                           dim3(BlockSize),
+                                           0,
+                                           arg.p_a_grid_,
+                                           arg.p_b_grid_,
+                                           arg.p_ds_grid_,
+                                           arg.p_e_grid_,
+                                           arg.a_grid_desc_k0_m_k1_,
+                                           arg.b_grid_desc_k0_n_k1_,
+                                           arg.ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                           arg.e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                           arg.a_element_op_,
+                                           arg.b_element_op_,
+                                           arg.cde_element_op_,
+                                           arg.block_2_ctile_map_);
+            }
+
+            return ave_time;
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
+        {
+            if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+        // check vector load/store
+        {
+            using Row = ck::tensor_layout::gemm::RowMajor;
+            using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+            // check vector load of A
+            if constexpr(is_same_v<ALayout, Row> && ABlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Col> && ABlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.MRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector laod of B
+            if constexpr(is_same_v<BLayout, Col> && BBlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<BLayout, Row> && BBlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.NRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector load of Ds
+            // only support RowMajor for now
+            bool all_valid = true;
+
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                if constexpr(!is_same_v<DLayout, Row>)
+                {
+                    all_valid = false;
+                }
+            });
+
+            if(!all_valid)
+            {
+                return false;
+            }
+
+            // check vector store of E
+            // only support RowMajor for now
+            if constexpr(is_same_v<ELayout, Row>)
+            {
+                if(arg.NRaw_ % CDEShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+        }
+
+        return GridwiseOp::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                         arg.b_grid_desc_k0_n_k1_,
+                                         arg.ds_grid_desc_m_n_,
+                                         arg.e_grid_desc_m_n_,
+                                         arg.block_2_ctile_map_);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_a,
+                             const void* p_b,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t StrideA,
+                             index_t StrideB,
+                             std::array<ck::index_t, NumDTensor> StrideDs,
+                             index_t StrideE,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_e,
+                        M,
+                        N,
+                        K,
+                        StrideA,
+                        StrideB,
+                        StrideDs,
+                        StrideE,
+                        1,
+                        1,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        index_t M,
+                        index_t N,
+                        index_t K,
+                        index_t StrideA,
+                        index_t StrideB,
+                        std::array<ck::index_t, NumDTensor> StrideDs,
+                        index_t StrideE,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          M,
+                                          N,
+                                          K,
+                                          StrideA,
+                                          StrideB,
+                                          StrideDs,
+                                          StrideE,
+                                          1,
+                                          1,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<LoopScheduler, std::string> LoopSchedToString{
+            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceGemmMultipleD_Wmma_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerWMMA << ", "
+            << NPerWMMA << ", "
+            << MRepeat << ", "
+            << NRepeat
+            << ">"
+            << " NumPrefetch: "
+            << NumPrefetch << ", "
+            << "LoopScheduler: "
+            << LoopSchedToString[LoopSched] << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index 3830e1c0e822a7bb86622c860f623978cf0c6d53..d4ecea7bb7a815b9f5f98e71b4350c8f1ca7c43b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -51,7 +51,8 @@ __global__ void
                                                 e_grid_desc_mblock_mperblock_nblock_nperblock,
                                             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -431,9 +432,6 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
             const index_t grid_size =
                 arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
 
-            const auto K =
-                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
-
             auto launch_kernel = [&](auto has_main_k_block_loop) {
                 constexpr bool has_main_loop = has_main_k_block_loop.value;
 
@@ -471,6 +469,8 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
                                               arg.block_2_etile_map_);
             };
 
+            const auto K = arg.a_grid_desc_m_k_.GetLength(I1);
+
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
                 return launch_kernel(integral_constant<bool, true>{});
@@ -491,7 +491,9 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
@@ -681,6 +683,14 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
             << KPerBlock << ", "
             << AK1 << ", "
             << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
             << getGemmSpecializationString(GemmSpec)
             << ">"
             << " LoopScheduler: "
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
index 3f62601f966b2e9a9ff7503283e54729644df72e..e60fae370f7b4932b7952dbbe3dced5caca4a824 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -822,7 +822,15 @@ struct DeviceGemmReduce_Xdl_CShuffle : public DeviceGemmReduce<0, ReduceOperatio
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle 
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..f64450b75f272dbb0f9ef4066e055a2f59c3879e
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
@@ -0,0 +1,606 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t K0PerBlock,
+          ck::index_t K1,
+          ck::index_t MPerWMMA,
+          ck::index_t NPerWMMA,
+          ck::index_t MRepeat,
+          ck::index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsAddExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsAddExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
+          ck::index_t NumPrefetch         = 1,
+          ck::LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
+struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
+                                                   BLayout,
+                                                   CLayout,
+                                                   ADataType,
+                                                   BDataType,
+                                                   CDataType,
+                                                   AElementwiseOperation,
+                                                   BElementwiseOperation,
+                                                   CElementwiseOperation>
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    // K1 = Max Vector Access Pixels
+    static constexpr auto K1Number = Number<K1>{};
+
+    static constexpr auto matrix_padder =
+        MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, K0PerBlock* K1};
+
+    static auto MakeAGridDescriptor_K0_M_K1(index_t MRaw, index_t KRaw, index_t StrideA)
+    {
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto a_grid_desc_m_k = matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+        const auto M               = a_grid_desc_m_k.GetLength(I0);
+        const auto K               = a_grid_desc_m_k.GetLength(I1);
+        assert(K % K1 == 0);
+        const index_t K0 = K / K1;
+
+        return transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                       make_pass_through_transform(M)),
+            make_tuple(Sequence<1>{}, Sequence<0>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    static auto MakeBGridDescriptor_K0_N_K1(index_t KRaw, index_t NRaw, index_t StrideB)
+    {
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(StrideB, I1));
+            }
+        }();
+
+        const auto b_grid_desc_n_k = matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+        const auto N               = b_grid_desc_n_k.GetLength(I0);
+        const auto K               = b_grid_desc_n_k.GetLength(I1);
+        assert(K % K1 == 0);
+        const index_t K0 = K / K1;
+
+        return transform_tensor_descriptor(
+            b_grid_desc_n_k,
+            make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                       make_pass_through_transform(N)),
+            make_tuple(Sequence<1>{}, Sequence<0>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    static auto MakeCGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideC)
+    {
+        const auto c_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideC));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(c_grid_desc_mraw_nraw);
+    }
+
+    // Gridwise descriptor, mapping to whole given provblem.
+    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
+    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
+    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_wmma<
+        BlockSize,
+        ADataType,
+        BDataType,
+        AccDataType,
+        CShuffleDataType,
+        CDataType,
+        InMemoryDataOperationEnum::Set,
+        AGridDesc_K0_M_K1,
+        BGridDesc_K0_N_K1,
+        CGridDesc_M_N,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerWMMA,
+        NPerWMMA,
+        K1,
+        MRepeat,
+        NRepeat,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CShuffleBlockTransferScalarPerVector_NPerBlock,
+        NumPrefetch,
+        LoopSched,
+        PipelineVer>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const ADataType* p_a_grid,
+                 const BDataType* p_b_grid,
+                 CDataType* p_c_grid,
+                 index_t M,
+                 index_t N,
+                 index_t K,
+                 index_t StrideA,
+                 index_t StrideB,
+                 index_t StrideC,
+                 index_t M01,
+                 index_t N01,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CElementwiseOperation c_element_op)
+            : p_a_grid_{p_a_grid},
+              p_b_grid_{p_b_grid},
+              p_c_grid_{p_c_grid},
+              a_grid_desc_k0_m_k1_{},
+              b_grid_desc_k0_n_k1_{},
+              c_grid_desc_m_n_{},
+              c_grid_desc_mblock_mperblock_nblock_nperblock{},
+              block_2_ctile_map_{},
+              M01_{M01},
+              N01_{N01},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              c_element_op_{c_element_op},
+              MRaw_{M},
+              NRaw_{N},
+              KRaw_{K}
+        {
+            a_grid_desc_k0_m_k1_ =
+                DeviceGemmWmma_CShuffle::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
+            b_grid_desc_k0_n_k1_ =
+                DeviceGemmWmma_CShuffle::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
+            c_grid_desc_m_n_ = DeviceGemmWmma_CShuffle::MakeCGridDescriptor_M_N(M, N, StrideC);
+
+            block_2_ctile_map_ =
+                GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_, M01, N01);
+
+            if(GridwiseGemm::CheckValidity(a_grid_desc_k0_m_k1_,
+                                           b_grid_desc_k0_n_k1_,
+                                           c_grid_desc_m_n_,
+                                           block_2_ctile_map_))
+            {
+                c_grid_desc_mblock_mperblock_nblock_nperblock =
+                    GridwiseGemm::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        c_grid_desc_m_n_);
+            }
+        }
+
+        //  private:
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        CDataType* p_c_grid_;
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+        CGridDesc_M_N c_grid_desc_m_n_;
+        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            c_grid_desc_mblock_mperblock_nblock_nperblock;
+        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
+        index_t M01_;
+        index_t N01_;
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CElementwiseOperation c_element_op_;
+        // for checking vector load/store
+        index_t MRaw_;
+        index_t NRaw_;
+        index_t KRaw_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceGemmWmma_CShuffle::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+#if 0
+            {
+                std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
+                          << ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
+                          << ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) 
+                          << ", " << arg.c_grid_desc_m_n_.GetLength(I1) << ", "
+                          << arg.c_grid_desc_m_n_.GetLength(I2) << "}" << std::endl;
+            }
+#endif
+
+            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                            arg.b_grid_desc_k0_n_k1_,
+                                            arg.c_grid_desc_m_n_,
+                                            arg.block_2_ctile_map_))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemm_k0mk1_k0nk1_m0nm1_wmma_v1r1 has invalid setting");
+            }
+
+            const index_t grid_size =
+                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
+
+            const auto K =
+                arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
+
+            float ave_time = 0;
+
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            {
+                const auto kernel = kernel_gemm_wmma<
+                    GridwiseGemm,
+                    ADataType,
+                    BDataType,
+                    CDataType,
+                    remove_reference_t<DeviceGemmWmma_CShuffle::AGridDesc_K0_M_K1>,
+                    remove_reference_t<DeviceGemmWmma_CShuffle::BGridDesc_K0_N_K1>,
+                    remove_reference_t<
+                        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CElementwiseOperation,
+                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
+                    true>; // Last Option is W/O
+
+                ave_time = launch_and_time_kernel(stream_config,
+                                                  kernel,
+                                                  dim3(grid_size),
+                                                  dim3(BlockSize),
+                                                  0,
+                                                  arg.p_a_grid_,
+                                                  arg.p_b_grid_,
+                                                  arg.p_c_grid_,
+                                                  arg.a_grid_desc_k0_m_k1_,
+                                                  arg.b_grid_desc_k0_n_k1_,
+                                                  arg.c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  arg.a_element_op_,
+                                                  arg.b_element_op_,
+                                                  arg.c_element_op_,
+                                                  arg.block_2_ctile_map_);
+            }
+            else
+            {
+                const auto kernel = kernel_gemm_wmma<
+                    GridwiseGemm,
+                    ADataType,
+                    BDataType,
+                    CDataType,
+                    remove_reference_t<DeviceGemmWmma_CShuffle::AGridDesc_K0_M_K1>,
+                    remove_reference_t<DeviceGemmWmma_CShuffle::BGridDesc_K0_N_K1>,
+                    remove_reference_t<
+                        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CElementwiseOperation,
+                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
+                    false>;
+
+                ave_time = launch_and_time_kernel(stream_config,
+                                                  kernel,
+                                                  dim3(grid_size),
+                                                  dim3(BlockSize),
+                                                  0,
+                                                  arg.p_a_grid_,
+                                                  arg.p_b_grid_,
+                                                  arg.p_c_grid_,
+                                                  arg.a_grid_desc_k0_m_k1_,
+                                                  arg.b_grid_desc_k0_n_k1_,
+                                                  arg.c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  arg.a_element_op_,
+                                                  arg.b_element_op_,
+                                                  arg.c_element_op_,
+                                                  arg.block_2_ctile_map_);
+            }
+
+            return ave_time;
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
+        {
+            if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // check vector load/store
+        {
+            using Row = ck::tensor_layout::gemm::RowMajor;
+            using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+            // check vector load of A
+            if constexpr(is_same_v<ALayout, Row> && ABlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Col> && ABlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.MRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector laod of B
+            if constexpr(is_same_v<BLayout, Col> && BBlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<BLayout, Row> && BBlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.NRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector store of C
+            // only support RowMajor for now
+            if constexpr(is_same_v<CLayout, Row>)
+            {
+                if(arg.NRaw_ % CShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+        }
+
+        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                           arg.b_grid_desc_k0_n_k1_,
+                                           arg.c_grid_desc_m_n_,
+                                           arg.block_2_ctile_map_);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const ADataType* p_a,
+                             const BDataType* p_b,
+                             CDataType* p_c,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t StrideA,
+                             index_t StrideB,
+                             index_t StrideC,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CElementwiseOperation c_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_c,
+                        M,
+                        N,
+                        K,
+                        StrideA,
+                        StrideB,
+                        StrideC,
+                        1,
+                        1,
+                        a_element_op,
+                        b_element_op,
+                        c_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
+                                                      const void* p_b,
+                                                      void* p_c,
+                                                      index_t M,
+                                                      index_t N,
+                                                      index_t K,
+                                                      index_t StrideA,
+                                                      index_t StrideB,
+                                                      index_t StrideC,
+                                                      AElementwiseOperation a_element_op,
+                                                      BElementwiseOperation b_element_op,
+                                                      CElementwiseOperation c_element_op) override
+    {
+        return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
+                                          static_cast<const BDataType*>(p_b),
+                                          static_cast<CDataType*>(p_c),
+                                          M,
+                                          N,
+                                          K,
+                                          StrideA,
+                                          StrideB,
+                                          StrideC,
+                                          1,
+                                          1,
+                                          a_element_op,
+                                          b_element_op,
+                                          c_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<LoopScheduler, std::string> LoopSchedToString{
+            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceGemmWmma_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerWMMA << ", "
+            << NPerWMMA << ", "
+            << MRepeat << ", "
+            << NRepeat
+            << ">"
+            << " NumPrefetch: "
+            << NumPrefetch << ", "
+            << "LoopScheduler: "
+            << LoopSchedToString[LoopSched] << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
index ec6792cfaf83f4aa28458f7373384e24580ca240..ebbb1d38c000ba44ce53f9a8110114c41a299b14 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -75,136 +75,20 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
 
     static constexpr auto K1Number = Number<K1>{};
 
-    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
-    {
-        assert(K % K1 == 0);
-
-        const index_t K0 = K / K1;
-
-        const auto a_grid_desc_m_k = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-
-            return transform_tensor_descriptor(
-                a_grid_desc_m_k,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_right_pad_transform(M, PadM)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                a_grid_desc_m_k,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_pass_through_transform(M)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-    }
-
-    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
-    {
-        assert(K % K1 == 0);
-
-        const index_t K0 = K / K1;
-
-        const auto b_grid_desc_k_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-            return transform_tensor_descriptor(
-                b_grid_desc_k_n,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                b_grid_desc_k_n,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
-    {
-        const auto c_grid_desc_m_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-    }
-
-    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
-    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
-    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
     // GridwiseGemm
-    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<
+    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext<
         BlockSize,
         ADataType, // TODO: distinguish A/B datatype
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
+        ALayout,
+        BLayout,
+        CLayout,
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -236,173 +120,41 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
         LoopSched,
         PipelineVer>;
 
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t M,
-                 index_t N,
-                 index_t K,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 index_t M01,
-                 index_t N01,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              a_grid_desc_k0_m_k1_{},
-              b_grid_desc_k0_n_k1_{},
-              c_grid_desc_m_n_{},
-              c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
-              block_2_ctile_map_{},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              kraw_{K}
-        {
-            a_grid_desc_k0_m_k1_ = DeviceGemmXdl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
-            b_grid_desc_k0_n_k1_ = DeviceGemmXdl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
-            c_grid_desc_m_n_     = DeviceGemmXdl::MakeCGridDescriptor_M_N(M, N, StrideC);
-
-            block_2_ctile_map_ =
-                GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_, M01, N01);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_k0_m_k1_,
-                                           b_grid_desc_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
-                    GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
-            }
-        }
-
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
-        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t kraw_;
-    };
+    using Argument = typename GridwiseGemm::Argument;
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceGemmXdl::Argument;
-
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
         {
-#if DEBUG_LOG
+            if(stream_config.log_level_ > 0)
             {
-                std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
-                          << ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
-                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
-                          << ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
-                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                          << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+                karg.Print();
             }
-#endif
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                            arg.b_grid_desc_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(karg))
             {
                 throw std::runtime_error(
-                    "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3 has invalid setting");
+                    "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
-
-            const auto K =
-                arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
+            const auto [gdx, gdy, gdz] = GridwiseGemm::CalculateGridSize(karg.M, karg.N);
 
             float ave_time = 0;
 
-            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(karg.K))
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    true>;
-
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.block_2_ctile_map_);
+                const auto kernel = kernel_gemm_xdlops_v2r3<GridwiseGemm, true>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
             else
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    false>;
-
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.block_2_ctile_map_);
+                const auto kernel = kernel_gemm_xdlops_v2r3<GridwiseGemm, false>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
 
             return ave_time;
@@ -422,7 +174,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
         return true;
     }
 
-    static bool IsSupportedArgument(const Argument& arg)
+    static bool IsSupportedArgument(const Argument& karg)
     {
         if(ck::get_device_name() == "gfx908")
         {
@@ -432,7 +184,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                 return false;
             }
         }
-        else if(ck::get_device_name() == "gfx90a")
+        else if(ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx940")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
@@ -445,15 +197,12 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
             return false;
         }
 
-        if(arg.kraw_ % K1 != 0)
+        if(karg.K % K1 != 0)
         {
             return false;
         }
 
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                           arg.b_grid_desc_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(karg);
     }
 
     // polymorphic
@@ -471,24 +220,11 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation)
     {
-        return Argument{p_a,
-                        p_b,
-                        p_c,
-                        M,
-                        N,
-                        K,
-                        StrideA,
-                        StrideB,
-                        StrideC,
-                        1,
-                        1,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+        return Argument{p_a, p_b, p_c, M, N, K, StrideA, StrideB, StrideC};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -503,9 +239,9 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op) override
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
@@ -515,12 +251,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                                           K,
                                           StrideA,
                                           StrideB,
-                                          StrideC,
-                                          1,
-                                          1,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          StrideC);
     }
 
     // polymorphic
@@ -551,7 +282,11 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
             << MPerXDL << ", "
             << NPerXDL << ", "
             << MXdlPerWave << ", "
-            << NXdlPerWave
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
             << ">"
             << " NumPrefetch: "
             << NumPrefetch << ", "
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index 9ec8186d32ad8ab8f565981b36a17bba0734a1c0..b68ecfe7b14d1bfe632db55aa00abbbca68a8e84 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -82,276 +82,11 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
 
-    static auto MakeAGridDescriptor_AK0_M_AK1(index_t MRaw, index_t KRaw, index_t StrideA)
-    {
-        const auto a_grid_desc_mraw_kraw = [&]() {
-            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both M and K
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(M)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad M, but not K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_right_pad_transform(MRaw, MPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad K, but not M
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k = transform_tensor_descriptor(
-                a_grid_desc_mraw_kraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else
-        {
-            // not pad M or K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-    }
-
-    static auto MakeBGridDescriptor_BK0_N_BK1(index_t KRaw, index_t NRaw, index_t StrideB)
-    {
-        const auto b_grid_desc_nraw_kraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(I1, StrideB));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(StrideB, I1));
-            }
-        }();
-
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto NPad = N - NRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both N and K
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_right_pad_transform(NRaw, NPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(N)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad N, but not K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_right_pad_transform(NRaw, NPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad K, but not N
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k = transform_tensor_descriptor(
-                b_grid_desc_nraw_kraw,
-                make_tuple(make_pass_through_transform(NRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else
-        {
-            // not pad N or K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideC)
-    {
-        const auto c_grid_desc_mraw_nraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(I1, StrideC));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto NPad = N - NRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad M and N
-            return transform_tensor_descriptor(c_grid_desc_mraw_nraw,
-                                               make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                          make_right_pad_transform(NRaw, NPad)),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad M, but not N
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_right_pad_transform(MRaw, MPad), make_pass_through_transform(NRaw)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad N, but not M
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(NRaw, NPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-            // not pad M or N
-            return c_grid_desc_mraw_nraw;
-        }
-    }
-
-    using AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1(1, 1, 1));
-    using BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1(1, 1, 1));
-    using CGridDesc_M_N       = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1<
+        ALayout,
+        BLayout,
+        CLayout,
         ADataType, // TODO: distinguish A/B datatype
         GemmAccDataType,
         CShuffleDataType,
@@ -359,10 +94,8 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_AK0_M_AK1,
-        BGridDesc_BK0_N_BK1,
-        CGridDesc_M_N,
         NumGemmKPrefetchStage,
         BlockSize,
         MPerBlock,
@@ -397,163 +130,43 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
         LoopSched,
         PipelineVer>;
 
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t MRaw,
-                 index_t NRaw,
-                 index_t KRaw,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              a_grid_desc_ak0_m_ak1_{DeviceOp::MakeAGridDescriptor_AK0_M_AK1(MRaw, KRaw, StrideA)},
-              b_grid_desc_bk0_n_bk1_{DeviceOp::MakeBGridDescriptor_BK0_N_BK1(KRaw, NRaw, StrideB)},
-              c_grid_desc_m_n_{DeviceOp::MakeCGridDescriptor_M_N(MRaw, NRaw, StrideC)},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_ctile_map_{GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_)},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              kraw_{KRaw}
-        {
-            if(GridwiseGemm::CheckValidity(a_grid_desc_ak0_m_ak1_,
-                                           b_grid_desc_bk0_n_bk1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        c_grid_desc_m_n_);
-            }
-        }
-
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
-        BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            c_grid_desc_mblock_mperblock_nblock_nperblock_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t kraw_;
-    };
+    using Argument = typename GridwiseGemm::Argument;
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceOp::Argument;
-
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
-#if DEBUG_LOG
+            if(stream_config.log_level_ > 0)
             {
-                std::cout << "arg.a_grid_desc_ak0_m_ak1_{"
-                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) << ", "
-                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I1) << ", "
-                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.b_grid_desc_bk0_n_bk1_{"
-                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I0) << ", "
-                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I1) << ", "
-                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                          << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+                arg.Print();
             }
-#endif
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                            arg.b_grid_desc_bk0_n_bk1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(arg))
             {
                 throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
+            index_t gdx, gdy, gdz;
+            std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(arg.M, arg.N);
 
-            const auto K =
-                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
+            const auto K = GridwiseGemm::CalculateAK0(arg.K) * AK1;
 
             float ave_time = 0;
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    true>;
-
-                ave_time =
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_,
-                                           arg.p_b_grid_,
-                                           arg.p_c_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                const auto kernel = kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, true>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, arg);
             }
             else
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    false>;
-                ave_time =
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_,
-                                           arg.p_b_grid_,
-                                           arg.p_c_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                const auto kernel = kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, false>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, arg);
             }
 
             return ave_time;
@@ -575,24 +188,22 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
 
-        if((arg.kraw_ % AK1 != 0 || arg.kraw_ % BK1 != 0) &&
-           !(GemmSpec == GemmSpecialization::MKPadding ||
-             GemmSpec == GemmSpecialization::NKPadding ||
-             GemmSpec == GemmSpecialization::MNKPadding ||
-             GemmSpec == GemmSpecialization::KPadding))
+        if((arg.K % AK1 != 0 || arg.K % BK1 != 0) && !(GemmSpec == GemmSpecialization::MKPadding ||
+                                                       GemmSpec == GemmSpecialization::NKPadding ||
+                                                       GemmSpec == GemmSpecialization::MNKPadding ||
+                                                       GemmSpec == GemmSpecialization::KPadding))
         {
             return false;
         }
 
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(arg);
     }
 
     // polymorphic
@@ -604,28 +215,17 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
     static auto MakeArgument(const ADataType* p_a,
                              const BDataType* p_b,
                              CDataType* p_c,
-                             index_t MRaw,
-                             index_t NRaw,
-                             index_t KRaw,
+                             index_t M,
+                             index_t N,
+                             index_t K,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation)
     {
-        return Argument{p_a,
-                        p_b,
-                        p_c,
-                        MRaw,
-                        NRaw,
-                        KRaw,
-                        StrideA,
-                        StrideB,
-                        StrideC,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+        return Argument{p_a, p_b, p_c, M, N, K, StrideA, StrideB, StrideC};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -634,28 +234,25 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
     std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
                                                       const void* p_b,
                                                       void* p_c,
-                                                      index_t MRaw,
-                                                      index_t NRaw,
-                                                      index_t KRaw,
+                                                      index_t M,
+                                                      index_t N,
+                                                      index_t K,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op) override
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
                                           static_cast<CDataType*>(p_c),
-                                          MRaw,
-                                          NRaw,
-                                          KRaw,
+                                          M,
+                                          N,
+                                          K,
                                           StrideA,
                                           StrideB,
-                                          StrideC,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          StrideC);
     }
 
     // polymorphic
@@ -683,7 +280,15 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">"
             << " LoopScheduler: "
             << LoopSchedToString[LoopSched] << ", "
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
index ed3e6900bd0e073a7bc1ae61ea696b6394e5015e..a2403f109c50da0c21bb6272798b5b91bb7ed97b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -648,7 +648,9 @@ struct DeviceGemmLayerNorm_Xdl_CShuffle : public BaseOperator
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
@@ -760,7 +762,15 @@ struct DeviceGemmLayerNorm_Xdl_CShuffle : public BaseOperator
             << NPerBlock << ", "
             << KPerBlock << ", "
             << AK1 << ", "
-            << BK1
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
index 36b01f677f080e4f2f515738af6171d6ed4dbbcd..ef5e267819c7fd8ca3868f689fd7bf16dd50fbb8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index c85b805f59970f3931341f684a7d686d0716c4a9..6c07a5c56ce94a06f5602c722af100fe0a8460d9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -73,157 +73,24 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     static constexpr auto I2 = Number<2>{};
     static constexpr auto I3 = Number<3>{};
 
-    static constexpr auto K1Number = Number<K1>{};
+    // TODO: should be exposed as Tparams.
+    static constexpr index_t NumGemmKPrefetchStage = 1;
+    static constexpr LoopScheduler LoopSched       = make_default_loop_scheduler();
+    static constexpr PipelineVersion PipelineVer   = PipelineVersion::v1;
 
-    static auto
-    MakeAGridDescriptor_KBatch_K0_M_K1(index_t M, index_t K, index_t StrideA, int KBatch, int KPad)
-    {
-        assert(KPad % (K1 * KBatch) == 0);
-
-        const index_t K0 = KPad / (K1 * KBatch);
-
-        const auto a_grid_desc_m_k = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
-            a_grid_desc_m_k,
-            make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-            return transform_tensor_descriptor(
-                a_grid_desc_m_kpad,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_right_pad_transform(M, PadM)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                a_grid_desc_m_kpad,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_pass_through_transform(M)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-    }
-
-    static auto
-    MakeBGridDescriptor_KBatch_K0_N_K1(index_t K, index_t N, index_t StrideB, int KBatch, int KPad)
-    {
-        assert(KPad % (K1 * KBatch) == 0);
-
-        const index_t K0 = KPad / (K1 * KBatch);
-
-        const auto b_grid_desc_k_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
-            }
-        }();
-
-        const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
-            b_grid_desc_k_n,
-            make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-            return transform_tensor_descriptor(
-                b_grid_desc_kpad_n,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                b_grid_desc_kpad_n,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
-    {
-        const auto c_grid_desc_m_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-    }
-
-    static auto GetKPad(index_t K, index_t KBatch)
-    {
-        const index_t K0   = math::integer_divide_ceil(K, K1 * K0PerBlock * KBatch) * K0PerBlock;
-        const index_t KPad = KBatch * K0 * K1;
-        return KPad;
-    }
-
-    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_KBatch_K0_M_K1(1, 1, 1, 1, 1));
-    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_KBatch_K0_N_K1(1, 1, 1, 1, 1));
-    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
-    // GridwiseGemm
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,
         ADataType, // TODO: distinguish A/B datatype
         AccDataType,
         CDataType,
-        InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
+        ALayout,
+        BLayout,
+        CLayout,
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
+        NumGemmKPrefetchStage,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -251,238 +118,72 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         CShuffleMRepeatPerShuffle,
         CShuffleNRepeatPerShuffle,
         CBlockTransferScalarPerVector_NWaveNPerXDL,
-        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
+        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        LoopSched,
+        PipelineVer>;
 
-    // GridwiseGemm
-    using GridwiseGemmAtomicAdd = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
-        BlockSize,
-        ADataType, // TODO: distinguish A/B datatype
-        AccDataType,
-        CDataType,
-        InMemoryDataOperationEnum::AtomicAdd,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
-        AElementwiseOperation,
-        BElementwiseOperation,
-        CElementwiseOperation,
-        MPerBlock,
-        NPerBlock,
-        K0PerBlock,
-        MPerXDL,
-        NPerXDL,
-        K1,
-        MXdlPerWave,
-        NXdlPerWave,
-        ABlockTransferThreadClusterLengths_K0_M_K1,
-        ABlockTransferThreadClusterArrangeOrder,
-        ABlockTransferSrcAccessOrder,
-        ABlockTransferSrcVectorDim,
-        ABlockTransferSrcScalarPerVector,
-        ABlockTransferDstScalarPerVector_K1,
-        false, // AThreadTransferSrcResetCoordinateAfterRun,
-        ABlockLdsAddExtraM,
-        BBlockTransferThreadClusterLengths_K0_N_K1,
-        BBlockTransferThreadClusterArrangeOrder,
-        BBlockTransferSrcAccessOrder,
-        BBlockTransferSrcVectorDim,
-        BBlockTransferSrcScalarPerVector,
-        BBlockTransferDstScalarPerVector_K1,
-        false, // BThreadTransferSrcResetCoordinateAfterRun,
-        BBlockLdsAddExtraN,
-        CShuffleMRepeatPerShuffle,
-        CShuffleNRepeatPerShuffle,
-        CBlockTransferScalarPerVector_NWaveNPerXDL,
-        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
-
-    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
-        decltype(GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}));
-
-    using Block2CTileMap = typename GridwiseGemm::CBlockClusterAdaptor;
-
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t M,
-                 index_t N,
-                 index_t K,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 index_t M01,
-                 index_t N01,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op,
-                 index_t k_batch)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              a_grid_desc_kbatch_k0_m_k1_{},
-              b_grid_desc_kbatch_k0_n_k1_{},
-              c_grid_desc_m_n_{},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_ctile_map_{},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              k_batch_{k_batch}
-        {
-            int KPad = DeviceGemmXdlSplitKCShuffle::GetKPad(K, k_batch_);
-
-            a_grid_desc_kbatch_k0_m_k1_ =
-                DeviceGemmXdlSplitKCShuffle::MakeAGridDescriptor_KBatch_K0_M_K1(
-                    M, K, StrideA, k_batch_, KPad);
-            b_grid_desc_kbatch_k0_n_k1_ =
-                DeviceGemmXdlSplitKCShuffle::MakeBGridDescriptor_KBatch_K0_N_K1(
-                    K, N, StrideB, k_batch_, KPad);
-            c_grid_desc_m_n_ = DeviceGemmXdlSplitKCShuffle::MakeCGridDescriptor_M_N(M, N, StrideC);
-
-            block_2_ctile_map_ =
-                GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_kbatch_k0_m_k1_,
-                                           b_grid_desc_kbatch_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n_);
-            }
-        }
-
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        AGridDesc_K0_M_K1 a_grid_desc_kbatch_k0_m_k1_;
-        BGridDesc_K0_N_K1 b_grid_desc_kbatch_k0_n_k1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock c_grid_desc_mblock_mperblock_nblock_nperblock_;
-        Block2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t k_batch_;
-    };
+    using Argument              = typename GridwiseGemm::Argument;
+    using DefaultBlock2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceGemmXdlSplitKCShuffle::Argument;
 
-        void Print(const Argument& arg)
-        {
-            std::cout << "arg.a_grid_desc_kbatch_k0_m_k1_{"
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I0) << ", "
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1) << ", "
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I2) << ", "
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I3) << "}" << std::endl;
-
-            std::cout << "arg.b_grid_desc_kbatch_k0_n_k1_{"
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I0) << ", "
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I1) << ", "
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I2) << ", "
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I3) << "}" << std::endl;
-
-            std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                      << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
-        }
+        void Print(const Argument& karg) { karg.Print(); }
 
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
         {
             if(stream_config.log_level_ > 0)
             {
-                Print(arg);
+                Print(karg);
             }
 
-            const auto kbatch = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I0);
+            const auto kbatch = karg.k_batch;
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
-                                            arg.b_grid_desc_kbatch_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(karg))
             {
                 throw std::runtime_error(
-                    "wrong! GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2 has invalid setting");
+                    "wrong! GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2 has invalid "
+                    "setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
-
-            const auto K0 = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1);
+            const auto b2c_map = DefaultBlock2CTileMap{};
+            index_t gdx, gdy, gdz;
+            std::tie(gdx, gdy, gdz) = b2c_map.CalculateGridSize(karg.M, karg.N, karg.k_batch);
+            const auto K0           = karg.K0;
 
             const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
 
             float ave_time = 0;
 
             const auto Run = [&](const auto& kernel) {
-                hipGetErrorString(hipMemset(
-                    arg.p_c_grid_,
-                    0,
-                    arg.c_grid_desc_mblock_mperblock_nblock_nperblock_.GetElementSpaceSize() *
-                        sizeof(CDataType)));
-
-                ave_time =
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_,
-                                           arg.p_b_grid_,
-                                           arg.p_c_grid_,
-                                           arg.a_grid_desc_kbatch_k0_m_k1_,
-                                           arg.b_grid_desc_kbatch_k0_n_k1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.block_2_ctile_map_);
+                if(kbatch > 1)
+                    hipGetErrorString(
+                        hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(CDataType)));
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg, b2c_map);
             };
 
             if(has_main_k0_block_loop)
             {
                 if(kbatch == 1)
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        true>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             true,
+                                                             InMemoryDataOperationEnum::Set,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemmAtomicAdd,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        true>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             true,
+                                                             InMemoryDataOperationEnum::AtomicAdd,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
@@ -491,37 +192,21 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
             {
                 if(kbatch == 1)
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        false>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             false,
+                                                             InMemoryDataOperationEnum::Set,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemmAtomicAdd,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        false>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             false,
+                                                             InMemoryDataOperationEnum::AtomicAdd,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
@@ -544,12 +229,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         return true;
     }
 
-    static bool IsSupportedArgument(const Argument& arg)
+    static bool IsSupportedArgument(const Argument& karg)
     {
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
-                                           arg.b_grid_desc_kbatch_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(karg);
     }
 
     // polymorphic
@@ -567,9 +249,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op,
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation,
                              index_t KBatch)
     {
         return Argument{p_a,
@@ -581,11 +263,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         StrideA,
                         StrideB,
                         StrideC,
-                        1,
-                        1,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op,
+                        GridwiseGemm::CalculateMPadded(M),
+                        GridwiseGemm::CalculateNPadded(N),
+                        GridwiseGemm::CalculateKPadded(K, KBatch),
+                        GridwiseGemm::CalculateK0(K, KBatch),
                         KBatch};
     }
 
@@ -601,9 +282,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op,
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation,
                                                       ck::index_t KBatch = 1) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
@@ -615,11 +296,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                           StrideA,
                                           StrideB,
                                           StrideC,
-                                          1,
-                                          1,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op,
+                                          GridwiseGemm::CalculateMPadded(M),
+                                          GridwiseGemm::CalculateNPadded(N),
+                                          GridwiseGemm::CalculateKPadded(K, KBatch),
+                                          GridwiseGemm::CalculateK0(K, KBatch),
                                           KBatch);
     }
 
@@ -630,22 +310,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     }
 
     // polymorphic
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "DeviceGemmXdlSplitKCShuffle"
-            << "<"
-            << BlockSize << ", "
-            << MPerBlock << ", "
-            << NPerBlock << ", "
-            << K0PerBlock
-            << ">";
-        // clang-format on
-
-        return str.str();
-    }
+    std::string GetTypeString() const override { return GridwiseGemm::GetTypeString(); }
 };
 
 } // namespace device
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
similarity index 62%
rename from include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
index 19140688283ba4137e0420f6dcaead446c015fe6..d0de101c49c563ae40086f91cfb1232b9c11e81d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -10,74 +10,67 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 
 namespace ck {
 
 template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatDsPointer,
-          typename FloatE,
+          typename ABDataType,
+          typename EDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
+          typename EElementwiseOperation,
           typename AGridDesc_AK0_M_AK1,
           typename BGridDesc_BK0_N_BK1,
-          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
           typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
           typename Block2ETileMap,
           bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+    __launch_bounds__(CK_WAVELET_MAX_THREAD_PER_BLOCK, CK_WAVELET_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_bias_e_permute(const FloatAB* __restrict__ p_a_grid,
-                                   const FloatAB* __restrict__ p_b_grid,
-                                   FloatDsPointer p_ds_grid,
-                                   FloatE* __restrict__ p_e_grid,
-                                   const AElementwiseOperation a_element_op,
-                                   const BElementwiseOperation b_element_op,
-                                   const CDEElementwiseOperation cde_element_op,
-                                   const AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1,
-                                   const BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1,
-                                   const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                                       ds_grid_desc_mblock_mperblock_nblock_nperblock,
-                                   const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                                       e_grid_desc_mblock_mperblock_nblock_nperblock,
-                                   const Block2ETileMap block_2_etile_map)
+        kernel_gemm_xdl_waveletmodel_cshuffle(
+            const ABDataType* __restrict__ p_a_grid,
+            const ABDataType* __restrict__ p_b_grid,
+            EDataType* __restrict__ p_e_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const EElementwiseOperation e_element_op,
+            const AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1,
+            const BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1,
+            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+            const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
                                                   p_b_grid,
-                                                  p_ds_grid,
                                                   p_e_grid,
                                                   p_shared,
                                                   a_element_op,
                                                   b_element_op,
-                                                  cde_element_op,
+                                                  e_element_op,
                                                   a_grid_desc_ak0_m_ak1,
                                                   b_grid_desc_bk0_n_bk1,
-                                                  ds_grid_desc_mblock_mperblock_nblock_nperblock,
                                                   e_grid_desc_mblock_mperblock_nblock_nperblock,
                                                   block_2_etile_map);
 #else
     ignore = p_a_grid;
     ignore = p_b_grid;
-    ignore = p_ds_grid;
     ignore = p_e_grid;
     ignore = a_element_op;
     ignore = b_element_op;
-    ignore = cde_element_op;
+    ignore = e_element_op;
     ignore = a_grid_desc_ak0_m_ak1;
     ignore = b_grid_desc_bk0_n_bk1;
-    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
     ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
     ignore = block_2_etile_map;
 #endif
@@ -89,27 +82,21 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-// input : A[M, K], or A[K, N]
-// input : B[K, N], or A[N, K]
-// input : D0[M, N], D1[M, N], ...
-// output : E[M, N]
-// C = a_op(A) * b_op(B)
-// E = cde_op(C, D0, D1, ...)
 template <typename ALayout,
           typename BLayout,
-          typename CDELayout,
+          typename ELayout,
           typename ADataType,
           typename BDataType,
-          typename AccDataType,
+          typename GemmAcEDataType,
           typename CShuffleDataType,
-          typename DDataType,
           typename EDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CDEElementwiseOperation,
           GemmSpecialization GemmSpec,
           index_t NumGemmKPrefetchStage,
-          index_t BlockSize,
+          index_t TileLoadThreadGroupSize,
+          index_t TileMathThreadGroupSize,
           index_t MPerBlock,
           index_t NPerBlock,
           index_t KPerBlock,
@@ -125,35 +112,37 @@ template <typename ALayout,
           index_t ABlockTransferSrcVectorDim,
           index_t ABlockTransferSrcScalarPerVector,
           index_t ABlockTransferDstScalarPerVector_AK1,
-          index_t ABlockLdsExtraM,
+          bool ABlockLdsExtraM,
           typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
           typename BBlockTransferThreadClusterArrangeOrder,
           typename BBlockTransferSrcAccessOrder,
           index_t BBlockTransferSrcVectorDim,
           index_t BBlockTransferSrcScalarPerVector,
           index_t BBlockTransferDstScalarPerVector_BK1,
-          index_t BBlockLdsExtraN,
+          bool BBlockLdsExtraN,
           index_t CShuffleMXdlPerWavePerShuffle,
           index_t CShuffleNXdlPerWavePerShuffle,
-          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-          index_t CDEBlockTransferScalarPerVector_NPerBlock,
-          LoopScheduler LoopSched = make_default_loop_scheduler()>
-struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOperation,
-                                                                  BElementwiseOperation,
-                                                                  CDEElementwiseOperation>
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
+struct DeviceGemm_Xdl_WaveletModel_CShuffle : public DeviceGemm<ALayout,
+                                                                BLayout,
+                                                                ELayout,
+                                                                ADataType,
+                                                                BDataType,
+                                                                EDataType,
+                                                                AElementwiseOperation,
+                                                                BElementwiseOperation,
+                                                                CDEElementwiseOperation>
 {
-    using DeviceOp = DeviceGemmBiasEPermute_Xdl;
+    using DeviceOp = DeviceGemm_Xdl_WaveletModel_CShuffle;
 
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
 
     static constexpr auto matrix_padder =
         MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock};
 
-    static constexpr index_t NumDTensor = 1;
-
     static auto MakeAGridDescriptor_M_K(index_t MRaw, index_t KRaw, index_t StrideA)
     {
         const auto a_grid_desc_mraw_kraw = [&]() {
@@ -190,31 +179,20 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
         return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
     }
 
-    static auto MakeEGridDescriptor_M_N(DEGridDesc_M0_M1_M2_N0_N1 d_e_grid_desc)
+    template <typename ELay>
+    static auto MakeEGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideE)
     {
-        index_t M0 = d_e_grid_desc.M0_;
-        index_t M1 = d_e_grid_desc.M1_;
-        index_t M2 = d_e_grid_desc.M2_;
-        index_t N0 = d_e_grid_desc.N0_;
-        index_t N1 = d_e_grid_desc.N1_;
-
-        index_t stride_M0 = d_e_grid_desc.stride_M0_;
-        index_t stride_M1 = d_e_grid_desc.stride_M1_;
-        index_t stride_M2 = d_e_grid_desc.stride_M2_;
-        index_t stride_N0 = d_e_grid_desc.stride_N0_;
-        index_t stride_N1 = d_e_grid_desc.stride_N1_;
-
         const auto e_grid_desc_mraw_nraw = [&]() {
-            const auto e_grid_desc_m0_m1_m2_n0_n1 = make_naive_tensor_descriptor(
-                make_tuple(M0, M1, M2, N0, N1),
-                make_tuple(stride_M0, stride_M1, stride_M2, stride_N0, stride_N1));
-
-            return transform_tensor_descriptor(
-                e_grid_desc_m0_m1_m2_n0_n1,
-                make_tuple(make_merge_transform(make_tuple(M0, M1, M2)),
-                           make_merge_transform(make_tuple(N0, N1))),
-                make_tuple(Sequence<0, 1, 2>{}, Sequence<3, 4>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideE));
+            }
         }();
 
         return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
@@ -222,23 +200,24 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
 
     using AGridDesc_M_K = decltype(MakeAGridDescriptor_M_K(1, 1, 1));
     using BGridDesc_N_K = decltype(MakeBGridDescriptor_N_K(1, 1, 1));
-    using EGridDesc_M_N = decltype(MakeEGridDescriptor_M_N(DEGridDesc_M0_M1_M2_N0_N1{}));
-
-    using DsGridDesc_M_N = Tuple<EGridDesc_M_N>;
+    using EGridDesc_M_N = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
 
     // GridwiseGemm
-    using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
+    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdl_waveletmodel_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
-        AccDataType,
+        GemmAcEDataType,
         CShuffleDataType,
-        ck::Tuple<DDataType>,
         EDataType,
         AElementwiseOperation,
         BElementwiseOperation,
         CDEElementwiseOperation,
         InMemoryDataOperationEnum::Set,
+        AGridDesc_M_K,
+        BGridDesc_N_K,
+        EGridDesc_M_N,
         NumGemmKPrefetchStage,
-        BlockSize,
+        TileLoadThreadGroupSize,
+        TileMathThreadGroupSize,
         MPerBlock,
         NPerBlock,
         KPerBlock,
@@ -266,9 +245,8 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
         BBlockLdsExtraN,
         CShuffleMXdlPerWavePerShuffle,
         CShuffleNXdlPerWavePerShuffle,
-        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-        CDEBlockTransferScalarPerVector_NPerBlock,
-        LoopSched>;
+        CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CShuffleBlockTransferScalarPerVector_NPerBlock>;
 
     using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
         GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
@@ -280,91 +258,64 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
     // Argument
     struct Argument : public BaseArgument
     {
-        Argument(const void* p_a_grid,
-                 const void* p_b_grid,
-                 const void* p_d_grid,
-                 void* p_e_grid,
+        Argument(const ADataType* p_a_grid,
+                 const BDataType* p_b_grid,
+                 EDataType* p_e_grid,
                  index_t MRaw,
                  index_t NRaw,
                  index_t KRaw,
                  index_t StrideA,
                  index_t StrideB,
-                 DEGridDesc_M0_M1_M2_N0_N1 d_grid_desc,
-                 DEGridDesc_M0_M1_M2_N0_N1 e_grid_desc,
+                 index_t StrideE,
                  AElementwiseOperation a_element_op,
                  BElementwiseOperation b_element_op,
                  CDEElementwiseOperation cde_element_op)
             : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
               p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
-              p_ds_grid_{},
               p_e_grid_{static_cast<EDataType*>(p_e_grid)},
               a_grid_desc_m_k_{DeviceOp::MakeAGridDescriptor_M_K(MRaw, KRaw, StrideA)},
               b_grid_desc_n_k_{DeviceOp::MakeBGridDescriptor_N_K(KRaw, NRaw, StrideB)},
-              ds_grid_desc_m_n_{},
-              e_grid_desc_m_n_{DeviceOp::MakeEGridDescriptor_M_N(e_grid_desc)},
+              e_grid_desc_m_n_{DeviceOp::MakeEGridDescriptor_M_N<ELayout>(MRaw, NRaw, StrideE)},
               a_grid_desc_ak0_m_ak1_{
                   GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(a_grid_desc_m_k_)},
               b_grid_desc_bk0_n_bk1_{
                   GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(b_grid_desc_n_k_)},
-              ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
               e_grid_desc_mblock_mperblock_nblock_nperblock_{},
               block_2_etile_map_{GridwiseGemm::MakeDefaultBlock2ETileMap(e_grid_desc_m_n_)},
               a_element_op_{a_element_op},
               b_element_op_{b_element_op},
               cde_element_op_{cde_element_op}
         {
-
-            if(MRaw != d_grid_desc.M0_ * d_grid_desc.M1_ * d_grid_desc.M2_)
-            {
-                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
-            }
-
-            if(NRaw != d_grid_desc.N0_ * d_grid_desc.N1_)
-            {
-                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
-            }
-
-            // populate pointer, desc for Ds
-            // D pointer
-            p_ds_grid_(I0) = static_cast<const DDataType*>(p_d_grid);
-
-            // D desc
-            ds_grid_desc_m_n_(I0) = DeviceOp::MakeEGridDescriptor_M_N(d_grid_desc);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_m_k_,
-                                           b_grid_desc_n_k_,
-                                           ds_grid_desc_m_n_,
-                                           e_grid_desc_m_n_,
-                                           block_2_etile_map_))
+            if(GridwiseGemm::CheckValidity(
+                   a_grid_desc_m_k_, b_grid_desc_n_k_, e_grid_desc_m_n_, block_2_etile_map_))
             {
                 e_grid_desc_mblock_mperblock_nblock_nperblock_ =
                     GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
                         e_grid_desc_m_n_);
-
-                ds_grid_desc_mblock_mperblock_nblock_nperblock_(I0) =
-                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        ds_grid_desc_m_n_[I0]);
             }
         }
 
+        void Print() const
+        {
+            std::cout << "A[M, K]: " << a_grid_desc_m_k_ << std::endl;
+            std::cout << "B[N, K]: " << b_grid_desc_n_k_ << std::endl;
+            std::cout << "E[M, N]: " << e_grid_desc_m_n_ << std::endl;
+        }
+
         //  private:
         // pointers
         const ADataType* p_a_grid_;
         const BDataType* p_b_grid_;
-        typename GridwiseGemm::DsGridPointer p_ds_grid_;
         EDataType* p_e_grid_;
 
         // tensor descriptors for problem definiton
         AGridDesc_M_K a_grid_desc_m_k_;
         BGridDesc_N_K b_grid_desc_n_k_;
-        DsGridDesc_M_N ds_grid_desc_m_n_;
         EGridDesc_M_N e_grid_desc_m_n_;
 
         // tensor descriptors for block/thread-wise copy
         AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
         BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
-        typename GridwiseGemm::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
         typename GridwiseGemm::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
             e_grid_desc_mblock_mperblock_nblock_nperblock_;
 
@@ -384,56 +335,66 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
 
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
+#if 0
+            {
+                std::cout << "arg.a_grid_desc_ak0_m_ak1_{"
+                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) << ", "
+                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I1) << ", "
+                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.b_grid_desc_bk0_n_bk1_{"
+                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I0) << ", "
+                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I1) << ", "
+                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.e_grid_desc_m_n_{ " << arg.e_grid_desc_m_n_.GetLength(I0) << ", "
+                          << arg.e_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+            }
+#endif
+
             if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_m_k_,
                                             arg.b_grid_desc_n_k_,
-                                            arg.ds_grid_desc_m_n_,
                                             arg.e_grid_desc_m_n_,
                                             arg.block_2_etile_map_))
             {
                 throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
-
-            const auto K =
-                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
+            const index_t grid_size = GridwiseGemm::CalculateGridSize(arg.e_grid_desc_m_n_);
+            const auto K            = arg.a_grid_desc_m_k_.GetLength(I1);
 
             auto launch_kernel = [&](auto has_main_k_block_loop) {
                 constexpr bool has_main_loop = has_main_k_block_loop.value;
 
-                const auto kernel = kernel_gemm_bias_e_permute<
+                const auto kernel = kernel_gemm_xdl_waveletmodel_cshuffle<
                     GridwiseGemm,
                     ADataType, // TODO: distiguish A/B datatype
-                    typename GridwiseGemm::DsGridPointer,
                     EDataType,
                     AElementwiseOperation,
                     BElementwiseOperation,
                     CDEElementwiseOperation,
                     DeviceOp::AGridDesc_AK0_M_AK1,
                     DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
                     typename GridwiseGemm::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
                     typename GridwiseGemm::DefaultBlock2ETileMap,
                     has_main_loop>;
 
-                return launch_and_time_kernel(stream_config,
-                                              kernel,
-                                              dim3(grid_size),
-                                              dim3(BlockSize),
-                                              0,
-                                              arg.p_a_grid_,
-                                              arg.p_b_grid_,
-                                              arg.p_ds_grid_,
-                                              arg.p_e_grid_,
-                                              arg.a_element_op_,
-                                              arg.b_element_op_,
-                                              arg.cde_element_op_,
-                                              arg.a_grid_desc_ak0_m_ak1_,
-                                              arg.b_grid_desc_bk0_n_bk1_,
-                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                              arg.block_2_etile_map_);
+                return launch_and_time_kernel(
+                    stream_config,
+                    kernel,
+                    dim3(grid_size),
+                    dim3(TileLoadThreadGroupSize + TileMathThreadGroupSize),
+                    0,
+                    arg.p_a_grid_,
+                    arg.p_b_grid_,
+                    arg.p_e_grid_,
+                    arg.a_element_op_,
+                    arg.b_element_op_,
+                    arg.cde_element_op_,
+                    arg.a_grid_desc_ak0_m_ak1_,
+                    arg.b_grid_desc_bk0_n_bk1_,
+                    arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
+                    arg.block_2_etile_map_);
             };
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
@@ -456,14 +417,15 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
 
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_m_k_,
                                            arg.b_grid_desc_n_k_,
-                                           arg.ds_grid_desc_m_n_,
                                            arg.e_grid_desc_m_n_,
                                            arg.block_2_etile_map_);
     }
@@ -474,32 +436,28 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
         return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
     }
 
-    static auto MakeArgument(const void* p_a,
-                             const void* p_b,
-                             const void* p_d,
-                             void* p_e,
+    static auto MakeArgument(const ADataType* p_a,
+                             const BDataType* p_b,
+                             EDataType* p_e,
                              index_t MRaw,
                              index_t NRaw,
                              index_t KRaw,
                              index_t StrideA,
                              index_t StrideB,
-                             DEGridDesc_M0_M1_M2_N0_N1 d_grid_desc,
-                             DEGridDesc_M0_M1_M2_N0_N1 e_grid_desc,
+                             index_t StrideE,
                              AElementwiseOperation a_element_op,
                              BElementwiseOperation b_element_op,
                              CDEElementwiseOperation cde_element_op)
     {
         return Argument{p_a,
                         p_b,
-                        p_d,
                         p_e,
                         MRaw,
                         NRaw,
                         KRaw,
                         StrideA,
                         StrideB,
-                        d_grid_desc,
-                        e_grid_desc,
+                        StrideE,
                         a_element_op,
                         b_element_op,
                         cde_element_op};
@@ -511,30 +469,26 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_a,
                         const void* p_b,
-                        const void* p_d,
                         void* p_e,
                         index_t MRaw,
                         index_t NRaw,
                         index_t KRaw,
                         index_t StrideA,
                         index_t StrideB,
-                        DEGridDesc_M0_M1_M2_N0_N1 d_grid_desc,
-                        DEGridDesc_M0_M1_M2_N0_N1 e_grid_desc,
+                        index_t StrideE,
                         AElementwiseOperation a_element_op,
                         BElementwiseOperation b_element_op,
                         CDEElementwiseOperation cde_element_op) override
     {
-        return std::make_unique<Argument>(p_a,
-                                          p_b,
-                                          p_d,
-                                          p_e,
+        return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
+                                          static_cast<const BDataType*>(p_b),
+                                          static_cast<EDataType*>(p_e),
                                           MRaw,
                                           NRaw,
                                           KRaw,
                                           StrideA,
                                           StrideB,
-                                          d_grid_desc,
-                                          e_grid_desc,
+                                          StrideE,
                                           a_element_op,
                                           b_element_op,
                                           cde_element_op);
@@ -552,9 +506,10 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DeviceGemmBiasEPermute_Xdl"
+        str << "DeviceGemm_Xdl_WaveletModel_CShuffle"
             << "<"
-            << BlockSize << ", "
+	    << TileLoadThreadGroupSize << ", "
+            << TileMathThreadGroupSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
             << KPerBlock << ", "
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index 03d9e26a460962222c90f393df7115fcd4ecdf76..4b1cc65ba28ff0aff771c57b3080caede43e8d84 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -37,7 +37,8 @@ __global__ void
             const BElementwiseOperation b_element_op,
             const CDEElementwiseOperation cde_element_op)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
@@ -651,11 +652,12 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
                 }
             }
 
-            hipGetErrorString(hipMemcpy(arg.p_workspace_,
-                                        arg.contraction_multi_d_kernel_args_.data(),
-                                        arg.contraction_multi_d_kernel_args_.size() *
-                                            sizeof(ContractionMultiDKernelArg),
-                                        hipMemcpyHostToDevice));
+            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
+                                                  arg.contraction_multi_d_kernel_args_.data(),
+                                                  arg.contraction_multi_d_kernel_args_.size() *
+                                                      sizeof(ContractionMultiDKernelArg),
+                                                  hipMemcpyHostToDevice,
+                                                  stream_config.stream_id_));
 
             float ave_time = 0;
 
@@ -703,7 +705,9 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 682aba086008b607ac2f4313b6de45e66008bb63..95af932519c4977a7ffe773d01b33a9f30165665 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -130,7 +130,8 @@ __global__ void
             const Block2ETileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -458,7 +459,6 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
               p_ds_grid_{},
               p_e_grid_{static_cast<EDataType*>(p_e)},
               num_group_{a_g_n_k_wos_lengths[0]},
-              num_gemm_{},
               a_element_op_{a_element_op},
               b_element_op_{b_element_op},
               cde_element_op_{cde_element_op},
@@ -507,9 +507,6 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
             const auto YTilde = ConvStrideH / GcdStrideDilationH;
             const auto XTilde = ConvStrideW / GcdStrideDilationW;
 
-            // number of GEMM
-            num_gemm_ = YTilde * XTilde;
-
             for(index_t i_ytilde = 0; i_ytilde < YTilde; ++i_ytilde)
             {
                 for(index_t i_xtilde = 0; i_xtilde < XTilde; ++i_xtilde)
@@ -625,7 +622,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
         void Print() const
         {
-            for(index_t i = 0; i < num_gemm_; i++)
+            for(std::size_t i = 0; i < a_grid_desc_ak0_m_ak1_container_.size(); i++)
             {
                 std::cout << "a_grid_desc_ak0_m_ak1_container_"
                           << a_grid_desc_ak0_m_ak1_container_[i] << std::endl;
@@ -653,7 +650,6 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
         // tensor descriptor for problem definition
         index_t num_group_;
-        index_t num_gemm_;
         std::vector<AGridDesc_M_K> a_grid_desc_m_k_container_;
         std::vector<BGridDesc_N_K> b_grid_desc_n_k_container_;
         std::vector<DsGridDesc_M_N> ds_grid_desc_m_n_container_;
@@ -707,7 +703,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
             float ave_time = 0;
 
-            for(index_t i = 0; i < arg.num_gemm_; i++)
+            for(std::size_t i = 0; i < arg.a_grid_desc_ak0_m_ak1_container_.size(); i++)
             {
                 if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_m_k_container_[i],
                                                 arg.b_grid_desc_n_k_container_[i],
@@ -806,7 +802,8 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
         }
 
         // vector load for A matrix from global memory to LDS
-        if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK>)
+        if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK> ||
+                     is_same_v<ALayout, tensor_layout::convolution::NHWGK>)
         {
             if(!(ABlockTransferSrcVectorDim == 2 && ConvK % ABlockTransferSrcScalarPerVector == 0))
             {
@@ -861,7 +858,8 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
         }
 
         // vector store for E
-        if constexpr(is_same_v<ELayout, tensor_layout::convolution::GNHWC>)
+        if constexpr(is_same_v<ELayout, tensor_layout::convolution::GNHWC> ||
+                     is_same_v<ELayout, tensor_layout::convolution::NHWGC>)
         {
             // vector store C matrix into global memory
             if(!(ConvC % CDEBlockTransferScalarPerVector_NPerBlock == 0))
@@ -1003,7 +1001,15 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
             << KPerBlock << ", "
             << AK1 << ", "
             << BK1 << ", "
-            << getConvBackwardDataSpecializationString(ConvBackwardDataSpecialization)
+            << getConvBackwardDataSpecializationString(ConvBackwardDataSpecialization) << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
 
         return str.str();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
similarity index 72%
rename from include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
index d9e7b54cc8b14cb5ba2cd93d4dd295b6e3670f71..003a508f079a633a102418e265643ef25dc19dd0 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -13,7 +13,7 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 
@@ -50,35 +50,28 @@ struct ComputePtrOffsetOfStridedBatch
 template <typename GridwiseGemm,
           typename FloatAB,
           typename FloatC,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename AGridDesc_B_K0_M_K1,
-          typename BGridDesc_B_K0_N_K1,
-          typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename AGridDesc_B_K0_M0_M1_K1,
+          typename BGridDesc_B_K0_N0_N1_K1,
+          typename CGridDesc_M0_M10_M11_N0_N10_N11,
           typename Block2CTileMap,
           typename ComputePtrOffsetOfBatch,
-          bool HasMainKBlockLoop>
+          bool HasMainKBlockLoop,
+          bool HasDoubleTailKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_batched_gemm_xdlops_bwd_weight(
+        kernel_batched_gemm_dlops_bwd_weight(
             const FloatAB* __restrict__ p_a_grid,
             const FloatAB* __restrict__ p_b_grid,
             FloatC* __restrict__ p_c_grid,
-            const AElementwiseOperation a_element_op,
-            const BElementwiseOperation b_element_op,
-            const CElementwiseOperation c_element_op,
             const index_t batch_count,
-            const AGridDesc_B_K0_M_K1 a_b_k0_m_k1_grid_desc,
-            const BGridDesc_B_K0_N_K1 b_b_k0_n_k1_grid_desc,
-            const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
-                c_grid_desc_mblock_mperblock_nblock_nperblock,
+            const AGridDesc_B_K0_M0_M1_K1 a_grid_desc_kbatch_k0_m0_m1_k1,
+            const BGridDesc_B_K0_N0_N1_K1 b_grid_desc_kbatch_k0_n0_n1_k1,
+            const CGridDesc_M0_M10_M11_N0_N10_N11 c_grid_desc_m0_m10_m11_n0_n10_n11,
             const Block2CTileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -92,38 +85,19 @@ __global__ void
 
     __shared__ FloatAB p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB)];
 
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
-                                                  p_b_grid + b_batch_offset,
-                                                  p_c_grid + c_batch_offset,
-                                                  p_shared,
-                                                  a_b_k0_m_k1_grid_desc,
-                                                  b_b_k0_n_k1_grid_desc,
-                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  block_2_ctile_map);
-#else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_c_grid;
-    ignore = a_b_k0_m_k1_grid_desc;
-    ignore = b_b_k0_n_k1_grid_desc;
-    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = batch_count;
-    ignore = block_2_ctile_map;
-    ignore = compute_ptr_offset_of_batch;
-
-    compute_ptr_offset_of_batch.GetAPtrOffset(0);
-    compute_ptr_offset_of_batch.GetBPtrOffset(0);
-    compute_ptr_offset_of_batch.GetCPtrOffset(0);
-#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+    GridwiseGemm::template Run<HasMainKBlockLoop, HasDoubleTailKBlockLoop>(
+        p_a_grid + a_batch_offset,
+        p_b_grid + b_batch_offset,
+        p_c_grid + c_batch_offset,
+        p_shared,
+        a_grid_desc_kbatch_k0_m0_m1_k1,
+        b_grid_desc_kbatch_k0_n0_n1_k1,
+        c_grid_desc_m0_m10_m11_n0_n10_n11,
+        block_2_ctile_map,
+        integral_constant<bool, HasMainKBlockLoop>{},
+        integral_constant<bool, HasDoubleTailKBlockLoop>{});
 }
 
-// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
 template <ck::index_t NDimSpatial,
           typename InDataType,
           typename WeiDataType,
@@ -138,29 +112,29 @@ template <ck::index_t NDimSpatial,
           ck::index_t NPerBlock,
           ck::index_t K0PerBlock,
           ck::index_t K1,
-          ck::index_t MPerXdl,
-          ck::index_t NPerXdl,
-          ck::index_t MXdlPerWave,
-          ck::index_t NXdlPerWave,
-          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          index_t M1PerThread,
+          index_t N1PerThread,
+          index_t KPerThread,
+          typename M1N1ThreadClusterM1Xs,
+          typename M1N1ThreadClusterN1Xs,
+          typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
           typename ABlockTransferThreadClusterArrangeOrder,
           typename ABlockTransferSrcAccessOrder,
-          ck::index_t ABlockTransferSrcVectorDim,
-          ck::index_t ABlockTransferSrcScalarPerVector,
-          ck::index_t ABlockTransferDstScalarPerVector_K1,
-          bool ABlockLdsAddExtraM,
-          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
+          typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+          typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
           typename BBlockTransferThreadClusterArrangeOrder,
           typename BBlockTransferSrcAccessOrder,
-          ck::index_t BBlockTransferSrcVectorDim,
-          ck::index_t BBlockTransferSrcScalarPerVector,
-          ck::index_t BBlockTransferDstScalarPerVector_K1,
-          bool BBlockLdsAddExtraN,
-          index_t CShuffleMXdlPerWavePerShuffle,
-          index_t CShuffleNXdlPerWavePerShuffle,
-          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-          index_t CBlockTransferScalarPerVector_NWaveNPerXdl>
-struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
+          typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
+          typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector>
+struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     : public DeviceGroupedConvBwdWeight<
           NDimSpatial,
           ck::tuple_element_t<NDimSpatial - 1,
@@ -182,7 +156,7 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
           WeiElementwiseOperation,
           OutElementwiseOperation>
 {
-    using DeviceOp = DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle;
+    using DeviceOp = DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl;
 
     using ADataType = OutDataType;
     using BDataType = InDataType;
@@ -221,27 +195,27 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
         const index_t Wi            = input_spatial_lengths[0];
         const index_t Wo            = output_spatial_lengths[0];
         const index_t X             = filter_spatial_lengths[0];
-        const index_t ConvStrideW   = conv_filter_strides[0];
-        const index_t ConvDilationW = conv_filter_dilations[0];
         const index_t InLeftPadW    = input_left_pads[0];
         const index_t InRightPadW   = input_right_pads[0];
+        const index_t ConvStrideW   = conv_filter_strides[0];
+        const index_t ConvDilationW = conv_filter_dilations[0];
 
         const index_t GemmKTotal = N * Wo;
         const index_t GemmM      = K;
@@ -292,7 +266,7 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
-            // C: weight tensor
+            // C: weights tensor
             const auto wei_gemmm_gemmn_grid_desc =
                 make_naive_tensor_descriptor_packed(make_tuple(K, X * C));
 
@@ -369,21 +343,21 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
                               wei_gemmm_gemmn_grid_desc);
         }
-    }
 
+    } // function end
     template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -396,18 +370,18 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         const index_t Y = filter_spatial_lengths[0];
         const index_t X = filter_spatial_lengths[1];
 
-        const index_t ConvStrideH = conv_filter_strides[0];
-        const index_t ConvStrideW = conv_filter_strides[1];
-
-        const index_t ConvDilationH = conv_filter_dilations[0];
-        const index_t ConvDilationW = conv_filter_dilations[1];
-
         const index_t InLeftPadH = input_left_pads[0];
         const index_t InLeftPadW = input_left_pads[1];
 
         const index_t InRightPadH = input_right_pads[0];
         const index_t InRightPadW = input_right_pads[1];
 
+        const index_t ConvStrideH = conv_filter_strides[0];
+        const index_t ConvStrideW = conv_filter_strides[1];
+
+        const index_t ConvDilationH = conv_filter_dilations[0];
+        const index_t ConvDilationW = conv_filter_dilations[1];
+
         const index_t GemmKTotal = N * Ho * Wo;
         const index_t GemmM      = K;
         const index_t GemmN      = C * X * Y;
@@ -536,21 +510,22 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
                               wei_gemmm_gemmn_grid_desc);
         }
-    }
+
+    } // function end
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -566,14 +541,6 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         const index_t Y = filter_spatial_lengths[1];
         const index_t X = filter_spatial_lengths[2];
 
-        const index_t ConvStrideD = conv_filter_strides[0];
-        const index_t ConvStrideH = conv_filter_strides[1];
-        const index_t ConvStrideW = conv_filter_strides[2];
-
-        const index_t ConvDilationD = conv_filter_dilations[0];
-        const index_t ConvDilationH = conv_filter_dilations[1];
-        const index_t ConvDilationW = conv_filter_dilations[2];
-
         const index_t InLeftPadD = input_left_pads[0];
         const index_t InLeftPadH = input_left_pads[1];
         const index_t InLeftPadW = input_left_pads[2];
@@ -582,6 +549,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         const index_t InRightPadH = input_right_pads[1];
         const index_t InRightPadW = input_right_pads[2];
 
+        const index_t ConvStrideD = conv_filter_strides[0];
+        const index_t ConvStrideH = conv_filter_strides[1];
+        const index_t ConvStrideW = conv_filter_strides[2];
+
+        const index_t ConvDilationD = conv_filter_dilations[0];
+        const index_t ConvDilationH = conv_filter_dilations[1];
+        const index_t ConvDilationW = conv_filter_dilations[2];
+
         const index_t GemmKTotal = N * Do * Ho * Wo;
         const index_t GemmM      = K;
         const index_t GemmN      = C * Z * X * Y;
@@ -719,6 +694,7 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
                               wei_gemmm_gemmn_grid_desc);
         }
+
     } // function end
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
@@ -751,109 +727,55 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                                                                   1);
     }
 
-    // type convert descs
-    template <typename Desc_M0>
-    static auto PadDescriptor_M0_1d(Desc_M0 desc_m0, index_t gridSize, index_t blockSize)
-    {
-        const auto m0           = desc_m0.GetLength(I0);
-        const index_t loop_step = gridSize * blockSize * 4;
-        const auto pad          = math::integer_least_multiple(m0, loop_step) - m0;
-        const auto desc_m0_pad =
-            transform_tensor_descriptor(desc_m0,
-                                        make_tuple(make_right_pad_transform(m0, pad)),
-                                        make_tuple(Sequence<0>{}),
-                                        make_tuple(Sequence<0>{}));
-        return desc_m0_pad;
-    }
-
-    template <index_t Dim>
-    static auto MakeDescriptor_M0(const std::array<index_t, Dim>& shape,
-                                  const std::array<index_t, Dim>& stride,
-                                  index_t gridSize,
-                                  index_t blockSize)
-    {
-        auto tupleOfShape  = generate_tuple([&](auto I) { return shape[I]; }, Number<Dim>{});
-        auto tupleOfStride = generate_tuple([&](auto I) { return stride[I]; }, Number<Dim>{});
-
-        // nd desc - [s0, s1, s2, ...]
-        const auto desc = make_naive_tensor_descriptor(tupleOfShape, tupleOfStride);
-
-        // merge nd to 1d desc - [s0 * s1 * ...]
-        if constexpr(Dim > 1)
-        {
-            const auto desc_m0 = transform_tensor_descriptor(
-                desc,
-                make_tuple(make_merge_transform(tupleOfShape)),
-                make_tuple(generate_sequence_v2([&](auto I) { return I; }, Number<Dim>{})),
-                make_tuple(Sequence<0>{}));
-
-            return PadDescriptor_M0_1d(desc_m0, gridSize, blockSize);
-        }
-        else
-            return PadDescriptor_M0_1d(desc, gridSize, blockSize);
-    }
-
-    using GridDesc_M0 = decltype(MakeDescriptor_M0<1>({1}, {1}, 1, 1));
-
     using ABCGridDescs = decltype(GetABCGridDesc<NDimSpatial>());
 
-    using AGridDesc_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>;
-    using BGridDesc_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>;
-    using CGridDesc_M_N     = remove_cvref_t<decltype(ABCGridDescs{}[I2])>;
-
-    using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight<
-        BlockSize,
-        ADataType, // TODO: distinguish A/B datatype
-        AccDataType,
-        CDataType,
-        InMemoryDataOperationEnum::AtomicAdd,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
-        AElementwiseOperation,
-        BElementwiseOperation,
-        CElementwiseOperation,
-        MPerBlock,
-        NPerBlock,
-        K0PerBlock,
-        MPerXdl,
-        NPerXdl,
-        K1,
-        MXdlPerWave,
-        NXdlPerWave,
-        ABlockTransferThreadClusterLengths_K0_M_K1,
-        ABlockTransferThreadClusterArrangeOrder,
-        ABlockTransferSrcAccessOrder,
-        ABlockTransferSrcVectorDim,
-        ABlockTransferSrcScalarPerVector,
-        ABlockTransferDstScalarPerVector_K1,
-        false, // AThreadTransferSrcResetCoordinateAfterRun,
-        ABlockLdsAddExtraM,
-        ABlockLdsM1PerBlock,
-        ABlockLdsM0PerBlock,
-        ABlockLdsM1Padding,
-        BBlockTransferThreadClusterLengths_K0_N_K1,
-        BBlockTransferThreadClusterArrangeOrder,
-        BBlockTransferSrcAccessOrder,
-        BBlockTransferSrcVectorDim,
-        BBlockTransferSrcScalarPerVector,
-        BBlockTransferDstScalarPerVector_K1,
-        false, // BThreadTransferSrcResetCoordinateAfterRun,
-        BBlockLdsAddExtraN,
-        BBlockLdsN1PerBlock,
-        BBlockLdsN0PerBlock,
-        BBlockLdsN1Padding,
-        CShuffleMXdlPerWavePerShuffle,
-        CShuffleNXdlPerWavePerShuffle,
-        CBlockTransferScalarPerVector_NWaveNPerXdl,
-        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-        true,
-        true>;
+    using AGridDesc_B_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>;
+    using BGridDesc_B_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>;
+    using CGridDesc_M_N       = remove_cvref_t<decltype(ABCGridDescs{}[I2])>;
+
+    using GridwiseGemm =
+        GridwiseGemmDl_bkm_bkn_mn_v1r3<BlockSize,
+                                       ADataType,
+                                       AccDataType,
+                                       CDataType,
+                                       InMemoryDataOperationEnum::Set,
+                                       AGridDesc_B_K0_M_K1,
+                                       BGridDesc_B_K0_N_K1,
+                                       CGridDesc_M_N,
+                                       MPerBlock,
+                                       NPerBlock,
+                                       K0PerBlock,
+                                       K1,
+                                       M1PerThread,
+                                       N1PerThread,
+                                       KPerThread,
+                                       M1N1ThreadClusterM1Xs,
+                                       M1N1ThreadClusterN1Xs,
+                                       ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+                                       ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+                                       ABlockTransferThreadClusterArrangeOrder,
+                                       ABlockTransferSrcAccessOrder,
+                                       ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+                                       ABlockTransferSrcVectorTensorContiguousDimOrder,
+                                       ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+                                       BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+                                       BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+                                       BBlockTransferThreadClusterArrangeOrder,
+                                       BBlockTransferSrcAccessOrder,
+                                       BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+                                       BBlockTransferSrcVectorTensorContiguousDimOrder,
+                                       BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+                                       CThreadTransferSrcDstAccessOrder,
+                                       CThreadTransferSrcDstVectorDim,
+                                       CThreadTransferDstScalarPerVector>;
 
     // Argument
-    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
-        decltype(GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}));
-
+    using AGridDesc_B_K0_M0_M1_K1 =
+        decltype(GridwiseGemm::MakeAGridDescriptor_B_K0_M0_M1_K1(AGridDesc_B_K0_M_K1{}));
+    using BGridDesc_B_K0_N0_N1_K1 =
+        decltype(GridwiseGemm::MakeBGridDescriptor_B_K0_N0_N1_K1(BGridDesc_B_K0_N_K1{}));
+    using CGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(CGridDesc_M_N{}));
     using Block2CTileMap =
         decltype(GridwiseGemm::MakeCBlockClusterAdaptor(CGridDesc_M_N{}, 1, 1, 1));
 
@@ -862,19 +784,19 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         Argument(const InDataType* p_in_grid,
                  WeiDataType* p_wei_grid,
                  const OutDataType* p_out_grid,
-                 ck::index_t G,
-                 ck::index_t N,
-                 ck::index_t K,
-                 ck::index_t C,
-                 std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                 std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                 std::array<ck::index_t, NDimSpatial> input_left_pads,
-                 std::array<ck::index_t, NDimSpatial> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
+                 const ck::index_t G,
+                 const ck::index_t N,
+                 const ck::index_t K,
+                 const ck::index_t C,
+                 const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial + 3>& /*input_strides*/,
+                 const std::array<ck::index_t, NDimSpatial + 3>& /*output_strides*/,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                 const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                  InElementwiseOperation in_element_op,
                  WeiElementwiseOperation wei_element_op,
                  OutElementwiseOperation out_element_op,
@@ -885,21 +807,20 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
               a_grid_desc_kbatch_k0_m_k1_{},
               b_grid_desc_kbatch_k0_n_k1_{},
               c_grid_desc_m_n_{},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
               block_2_ctile_map_{},
               compute_ptr_offset_of_batch_{},
-              M01_{M01},
-              N01_{N01},
               a_element_op_{out_element_op},
-              b_element_op_{in_element_op},
-              c_element_op_{wei_element_op},
+              b_element_op_{wei_element_op},
+              c_element_op_{in_element_op},
               Conv_G_{G},
               Conv_N_{N},
               Conv_K_{K},
               Conv_C_{C},
-              output_spatial_lengths_{output_spatial_lengths},
+              input_spatial_lengths_{input_spatial_lengths},
               filter_spatial_lengths_{filter_spatial_lengths},
+              output_spatial_lengths_{output_spatial_lengths},
               conv_filter_strides_{conv_filter_strides},
+              conv_filter_dilations_{conv_filter_dilations},
               input_left_pads_{input_left_pads},
               input_right_pads_{input_right_pads},
               k_batch_{split_k}
@@ -922,6 +843,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
             b_grid_desc_kbatch_k0_n_k1_ = descs[I1];
             c_grid_desc_m_n_            = descs[I2];
 
+            a_grid_desc_kbatch_k0_m0_m1_k1_ =
+                GridwiseGemm::MakeAGridDescriptor_B_K0_M0_M1_K1(a_grid_desc_kbatch_k0_m_k1_);
+            b_grid_desc_kbatch_k0_n0_n1_k1_ =
+                GridwiseGemm::MakeBGridDescriptor_B_K0_N0_N1_K1(b_grid_desc_kbatch_k0_n_k1_);
+            c_grid_desc_m0_m10_m11_n0_n10_n11_ =
+                GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(c_grid_desc_m_n_);
+            ck::index_t M01 = 1;
+            ck::index_t N01 = 1;
             block_2_ctile_map_ =
                 GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
 
@@ -944,47 +873,44 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                                 end(filter_spatial_lengths),
                                 index_t{1},
                                 std::multiplies<>{});
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_kbatch_k0_m_k1_,
-                                           b_grid_desc_kbatch_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n_);
-            }
         }
 
         const ADataType* p_a_grid_;
         const BDataType* p_b_grid_;
         CDataType* p_c_grid_;
-        AGridDesc_K0_M_K1 a_grid_desc_kbatch_k0_m_k1_;
-        BGridDesc_K0_N_K1 b_grid_desc_kbatch_k0_n_k1_;
+
+        AGridDesc_B_K0_M_K1 a_grid_desc_kbatch_k0_m_k1_;
+        BGridDesc_B_K0_N_K1 b_grid_desc_kbatch_k0_n_k1_;
         CGridDesc_M_N c_grid_desc_m_n_;
-        CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock c_grid_desc_mblock_mperblock_nblock_nperblock_;
 
+        AGridDesc_B_K0_M0_M1_K1 a_grid_desc_kbatch_k0_m0_m1_k1_;
+        BGridDesc_B_K0_N0_N1_K1 b_grid_desc_kbatch_k0_n0_n1_k1_;
+        CGridDesc_M0_M10_M11_N0_N10_N11 c_grid_desc_m0_m10_m11_n0_n10_n11_;
+
+        // DefaultBlock2CTileMap block_2_ctile_map_;
         Block2CTileMap block_2_ctile_map_;
 
         // for computing batch offset
         ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
 
-        index_t M01_;
-        index_t N01_;
-
-        InElementwiseOperation a_element_op_;
-        OutElementwiseOperation b_element_op_;
-        WeiElementwiseOperation c_element_op_;
+        // element-wise op
+        OutElementwiseOperation a_element_op_;
+        WeiElementwiseOperation b_element_op_;
+        InElementwiseOperation c_element_op_;
 
         // for checking IsSupportedArgument()
-        index_t Conv_G_;
-        index_t Conv_N_;
-        index_t Conv_K_;
-        index_t Conv_C_;
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides_;
-        std::array<ck::index_t, NDimSpatial> input_left_pads_;
-        std::array<ck::index_t, NDimSpatial> input_right_pads_;
+        const index_t Conv_G_;
+        const index_t Conv_N_;
+        const index_t Conv_K_;
+        const index_t Conv_C_;
+
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides_;
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations_;
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads_;
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads_;
         index_t k_batch_;
     };
 
@@ -1007,44 +933,42 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                       << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I2) << ", "
                       << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I3) << "}" << std::endl;
 
-            std::cout << "arg.c_grid_desc_m_n_{" << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
+            std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
                       << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
         }
 
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
+
+            ShowInfo(arg);
+
             if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
                                             arg.b_grid_desc_kbatch_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+                                            arg.c_grid_desc_m_n_))
             {
                 throw std::runtime_error(
-                    "wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v3r1 has invalid setting");
+                    "wrong! GridwiseGemm GridwiseGemmDl_bkm_bkn_mn_v1r3 has invalid setting");
             }
 
             const index_t grid_size =
                 arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_) * arg.Conv_G_;
 
-            const auto K0 = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1);
-
-            const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
-
-            auto launch_kernel = [&](auto has_main_k_block_loop) {
-                constexpr bool has_main_loop = has_main_k_block_loop.value;
+            auto launch_kernel = [&](auto has_main_k_block_loop,
+                                     auto has_double_tail_k_block_loop) {
+                constexpr bool has_main_loop   = has_main_k_block_loop.value;
+                constexpr bool has_double_loop = has_double_tail_k_block_loop.value;
 
-                const auto kernel = kernel_batched_gemm_xdlops_bwd_weight<
+                const auto kernel = kernel_batched_gemm_dlops_bwd_weight<
                     GridwiseGemm,
                     ADataType, // TODO: distiguish A/B datatype
                     CDataType,
-                    OutElementwiseOperation,
-                    InElementwiseOperation,
-                    WeiElementwiseOperation,
-                    remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                    remove_reference_t<DeviceOp::CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    remove_reference_t<DeviceOp::AGridDesc_B_K0_M0_M1_K1>,
+                    remove_reference_t<DeviceOp::BGridDesc_B_K0_N0_N1_K1>,
+                    remove_reference_t<DeviceOp::CGridDesc_M0_M10_M11_N0_N10_N11>,
                     remove_reference_t<DeviceOp::Block2CTileMap>,
                     ComputePtrOffsetOfStridedBatch,
-                    has_main_loop>;
+                    has_main_loop,
+                    has_double_loop>;
 
                 return launch_and_time_kernel(stream_config,
                                               kernel,
@@ -1054,24 +978,38 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                                               arg.p_a_grid_,
                                               arg.p_b_grid_,
                                               arg.p_c_grid_,
-                                              arg.a_element_op_,
-                                              arg.b_element_op_,
-                                              arg.c_element_op_,
                                               arg.Conv_G_,
-                                              arg.a_grid_desc_kbatch_k0_m_k1_,
-                                              arg.b_grid_desc_kbatch_k0_n_k1_,
-                                              arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.a_grid_desc_kbatch_k0_m0_m1_k1_,
+                                              arg.b_grid_desc_kbatch_k0_n0_n1_k1_,
+                                              arg.c_grid_desc_m0_m10_m11_n0_n10_n11_,
                                               arg.block_2_ctile_map_,
                                               arg.compute_ptr_offset_of_batch_);
             };
 
-            if(has_main_k0_block_loop)
+            const auto K0                    = arg.a_grid_desc_kbatch_k0_m0_m1_k1_.GetLength(I1);
+            const bool has_main_k_block_loop = GridwiseGemm::CalculateHasMainKBlockLoop(K0);
+            const bool has_double_tail_k_block_loop =
+                GridwiseGemm::CalculateHasDoubleTailKBlockLoop(K0);
+
+            if(has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, true>{});
+            }
+            else if(has_main_k_block_loop && !has_double_tail_k_block_loop)
             {
-                return launch_kernel(integral_constant<bool, true>{});
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, false>{});
+            }
+            else if(!has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, true>{});
             }
             else
             {
-                return launch_kernel(integral_constant<bool, false>{});
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, false>{});
             }
         }
 
@@ -1090,6 +1028,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        // check device
+        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+             ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+             ck::get_device_name() == "gfx1102"))
+        {
+            return false;
+        }
+
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
@@ -1104,25 +1050,59 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
             }
         }
 
-        // vector load A/B matrix from global memory
-        if(!(ABlockTransferSrcVectorDim == 2 && BBlockTransferSrcVectorDim == 2 &&
-             arg.Conv_K_ % ABlockTransferSrcScalarPerVector == 0 &&
-             arg.Conv_C_ % BBlockTransferSrcScalarPerVector == 0))
+        // matrix A
         {
-            return false;
+            auto srcVectorLengths = ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1{};
+            if(srcVectorLengths[I2] != 1 || srcVectorLengths[I3] != 1)
+            {
+                return false;
+            }
+            if(K1 % srcVectorLengths[I4] != 0 || K0PerBlock % srcVectorLengths[I1] != 0)
+            {
+                return false;
+            }
+
+            const index_t K = arg.Conv_K_;
+
+            if(K % (srcVectorLengths[I1] * srcVectorLengths[I4]) != 0)
+            {
+                return false;
+            }
+        }
+
+        // matrix B
+        {
+            auto srcLoadLenghts   = BBlockTransferThreadSliceLengths_K0_N0_N1_K1{};
+            auto srcVectorLengths = BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1{};
+            if(srcVectorLengths[I1] != 1 || srcVectorLengths[I4] != 1)
+            {
+                return false;
+            }
+            if(srcLoadLenghts[I2] % srcVectorLengths[I2] != 0 ||
+               srcLoadLenghts[I3] % srcVectorLengths[I3] != 0)
+            {
+                return false;
+            }
+
+            const index_t C = arg.Conv_K_;
+
+            if(C % (srcVectorLengths[I2] * srcVectorLengths[I3]) != 0)
+            {
+                return false;
+            }
         }
 
         // vector store C matrix into global memory
-        if(!(arg.Conv_C_ % CBlockTransferScalarPerVector_NWaveNPerXdl == 0))
+        if(!(arg.Conv_C_ % CThreadTransferDstScalarPerVector == 0))
         {
+            std::cout << "Not surpport,because: arg.Conv_C_ % CThreadTransferDstScalarPerVector = "
+                      << arg.Conv_C_ % CThreadTransferDstScalarPerVector << std::endl;
             return false;
         }
 
         // Gridwise GEMM size
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
-                                           arg.b_grid_desc_kbatch_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(
+            arg.a_grid_desc_kbatch_k0_m_k1_, arg.b_grid_desc_kbatch_k0_n_k1_, arg.c_grid_desc_m_n_);
     }
 
     bool IsSupportedArgument(const BaseArgument* p_arg) override
@@ -1133,17 +1113,19 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
     static auto MakeArgument(const InDataType* p_in_grid,
                              WeiDataType* p_wei_grid,
                              const OutDataType* p_out_grid,
-                             ck::index_t G,
-                             ck::index_t N,
-                             ck::index_t K,
-                             ck::index_t C,
-                             std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                             std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                             std::array<ck::index_t, NDimSpatial> input_left_pads,
-                             std::array<ck::index_t, NDimSpatial> input_right_pads,
+                             const ck::index_t G,
+                             const ck::index_t N,
+                             const ck::index_t K,
+                             const ck::index_t C,
+                             const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                             const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                             const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                             const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                              InElementwiseOperation in_element_op,
                              WeiElementwiseOperation wei_element_op,
                              OutElementwiseOperation out_element_op,
@@ -1159,12 +1141,12 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                         input_spatial_lengths,
                         filter_spatial_lengths,
                         output_spatial_lengths,
+                        input_strides,
+                        output_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
                         input_right_pads,
-                        1,
-                        1,
                         in_element_op,
                         wei_element_op,
                         out_element_op,
@@ -1177,17 +1159,19 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
     MakeArgumentPointer(const void* p_in_grid,
                         void* p_wei_grid,
                         const void* p_out_grid,
-                        ck::index_t G,
-                        ck::index_t N,
-                        ck::index_t K,
-                        ck::index_t C,
-                        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                        std::array<ck::index_t, NDimSpatial> input_left_pads,
-                        std::array<ck::index_t, NDimSpatial> input_right_pads,
+                        const ck::index_t G,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t C,
+                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                         InElementwiseOperation in_element_op,
                         WeiElementwiseOperation wei_element_op,
                         OutElementwiseOperation out_element_op,
@@ -1203,12 +1187,12 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                                           input_spatial_lengths,
                                           filter_spatial_lengths,
                                           output_spatial_lengths,
+                                          input_strides,
+                                          output_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
                                           input_right_pads,
-                                          1,
-                                          1,
                                           in_element_op,
                                           wei_element_op,
                                           out_element_op,
@@ -1225,13 +1209,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle"
+        str << "DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl"
             << "<"
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
             << K0PerBlock << ", "
-            << getConvBackwardWeightSpecializationString(ConvBackwardWeightSpecialization)
+            << getConvBackwardWeightSpecializationString(ConvBackwardWeightSpecialization) << ", "
+            << K1
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..7475dc414da7af61e1276bc898399ceff6a35f39
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
@@ -0,0 +1,1519 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <numeric>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+namespace {
+
+struct ComputePtrOffsetOfStridedBatch
+{
+    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideA_);
+    }
+
+    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideB_);
+    }
+
+    __host__ __device__ constexpr long_index_t GetCPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideC_);
+    }
+
+    index_t BatchStrideA_;
+    index_t BatchStrideB_;
+    index_t BatchStrideC_;
+};
+
+} // namespace
+
+template <typename GridwiseGemm,
+          typename FloatAB,
+          typename FloatC,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          typename AGridDesc_B_K0_M_K1,
+          typename BGridDesc_B_K0_N_K1,
+          typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename Block2CTileMap,
+          typename ComputePtrOffsetOfBatch,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_batched_gemm_xdlops_bwd_weight(
+            const FloatAB* __restrict__ p_a_grid,
+            const FloatAB* __restrict__ p_b_grid,
+            FloatC* __restrict__ p_c_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CElementwiseOperation c_element_op,
+            const index_t batch_count,
+            const AGridDesc_B_K0_M_K1 a_b_k0_m_k1_grid_desc,
+            const BGridDesc_B_K0_N_K1 b_b_k0_n_k1_grid_desc,
+            const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+                c_grid_desc_mblock_mperblock_nblock_nperblock,
+            const Block2CTileMap block_2_ctile_map,
+            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    const index_t num_blocks_per_batch =
+        __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+    const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
+
+    const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
+    const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
+    const long_index_t c_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetCPtrOffset(g_idx)));
+
+    __shared__ FloatAB p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB)];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
+                                                  p_b_grid + b_batch_offset,
+                                                  p_c_grid + c_batch_offset,
+                                                  p_shared,
+                                                  a_b_k0_m_k1_grid_desc,
+                                                  b_b_k0_n_k1_grid_desc,
+                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  a_element_op,
+                                                  b_element_op,
+                                                  c_element_op,
+                                                  block_2_ctile_map);
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_c_grid;
+    ignore = a_b_k0_m_k1_grid_desc;
+    ignore = b_b_k0_n_k1_grid_desc;
+    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = c_element_op;
+    ignore = batch_count;
+    ignore = block_2_ctile_map;
+    ignore = compute_ptr_offset_of_batch;
+
+    compute_ptr_offset_of_batch.GetAPtrOffset(0);
+    compute_ptr_offset_of_batch.GetBPtrOffset(0);
+    compute_ptr_offset_of_batch.GetCPtrOffset(0);
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename AccDataType,
+          typename InElementwiseOperation,
+          typename WeiElementwiseOperation,
+          typename OutElementwiseOperation,
+          ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t K0PerBlock,
+          ck::index_t K1,
+          ck::index_t MPerXdl,
+          ck::index_t NPerXdl,
+          ck::index_t MXdlPerWave,
+          ck::index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsAddExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsAddExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CBlockTransferScalarPerVector_NWaveNPerXdl>
+struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
+    : public DeviceGroupedConvBwdWeight<NDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        OutDataType,
+                                        InElementwiseOperation,
+                                        WeiElementwiseOperation,
+                                        OutElementwiseOperation>
+{
+    using DeviceOp = DeviceGroupedConvBwdWeight_Xdl_CShuffle;
+
+    using ADataType = OutDataType;
+    using BDataType = InDataType;
+    using CDataType = WeiDataType;
+
+    using AElementwiseOperation = OutElementwiseOperation;
+    using BElementwiseOperation = InElementwiseOperation;
+    using CElementwiseOperation = WeiElementwiseOperation;
+
+    // TODO make A/B datatype different
+    using ABDataType = InDataType;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr auto K1Number     = Number<K1>{};
+    static constexpr auto GemmK1Number = K1Number;
+
+    // Bytes per 32 lds bank: 32 * 4 bytes
+    static constexpr auto BankLength = 128;
+    static constexpr auto ElePerBank = BankLength / sizeof(ADataType);
+
+    // M1 & M0
+    static constexpr auto ABlockLdsM1PerBlock = ElePerBank / K1;
+    static constexpr auto ABlockLdsM0PerBlock = MPerBlock / ABlockLdsM1PerBlock;
+    static constexpr auto ABlockLdsM1Padding  = 4;
+
+    // N1 & N0
+    static constexpr auto BBlockLdsN1PerBlock = ElePerBank / K1;
+    static constexpr auto BBlockLdsN0PerBlock = NPerBlock / BBlockLdsN1PerBlock;
+    static constexpr auto BBlockLdsN1Padding  = 4;
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
+    static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
+    {
+        using namespace ck;
+
+        const index_t Wi            = input_spatial_lengths[0];
+        const index_t Wo            = output_spatial_lengths[0];
+        const index_t X             = filter_spatial_lengths[0];
+        const index_t ConvStrideW   = conv_filter_strides[0];
+        const index_t ConvDilationW = conv_filter_dilations[0];
+        const index_t InLeftPadW    = input_left_pads[0];
+        const index_t InRightPadW   = input_right_pads[0];
+
+        const index_t GemmKTotal = N * Wo;
+        const index_t GemmM      = K;
+        const index_t GemmN      = C * X;
+
+        const index_t GemmKBatch = batch_k;
+        const index_t GemmK0 =
+            math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
+            K0PerBlock;
+        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // A: output tensor
+            const auto out_gemmktotal_gemmm_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Wo, K));
+
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_gemmktotal_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Wi, C));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+        else
+        {
+            const auto out_gemmktotal_gemmm_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Wo, K));
+            const auto in_n_wi_c_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N, Wi, C));
+
+            // A: output tensor
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_n_wip_c_grid_desc = transform_tensor_descriptor(
+                in_n_wi_c_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                           make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+            const auto in_n_x_wo_c_grid_desc = transform_tensor_descriptor(
+                in_n_wip_c_grid_desc,
+                make_tuple(
+                    make_pass_through_transform(N),
+                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
+                    make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}));
+
+            const auto in_gemmktotal_gemmn_grid_desc =
+                transform_tensor_descriptor(in_n_x_wo_c_grid_desc,
+                                            make_tuple(make_merge_transform(make_tuple(X, C)),
+                                                       make_merge_transform(make_tuple(N, Wo))),
+                                            make_tuple(Sequence<1, 3>{}, Sequence<0, 2>{}),
+                                            make_tuple(Sequence<1>{}, Sequence<0>{}));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+    }
+
+    template <ck::index_t NDim,
+              typename ck::enable_if<NDim == 2 &&
+                                         is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
+                                         is_same_v<OutLayout, tensor_layout::convolution::GNHWK>,
+                                     bool>::type = false>
+    static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
+    {
+        using namespace ck;
+
+        const index_t Hi = input_spatial_lengths[0];
+        const index_t Wi = input_spatial_lengths[1];
+
+        const index_t Ho = output_spatial_lengths[0];
+        const index_t Wo = output_spatial_lengths[1];
+
+        const index_t Y = filter_spatial_lengths[0];
+        const index_t X = filter_spatial_lengths[1];
+
+        const index_t ConvStrideH = conv_filter_strides[0];
+        const index_t ConvStrideW = conv_filter_strides[1];
+
+        const index_t ConvDilationH = conv_filter_dilations[0];
+        const index_t ConvDilationW = conv_filter_dilations[1];
+
+        const index_t InLeftPadH = input_left_pads[0];
+        const index_t InLeftPadW = input_left_pads[1];
+
+        const index_t InRightPadH = input_right_pads[0];
+        const index_t InRightPadW = input_right_pads[1];
+
+        const index_t GemmKTotal = N * Ho * Wo;
+        const index_t GemmM      = K;
+        const index_t GemmN      = C * X * Y;
+
+        const index_t GemmKBatch = batch_k;
+        const index_t GemmK0 =
+            math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
+            K0PerBlock;
+        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // A: output tensor
+            const auto out_gemmktotal_gemmm_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
+
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_gemmktotal_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Hi * Wi, C));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+        else
+        {
+            const auto out_gemmktotal_gemmm_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
+            const auto in_n_hi_wi_c_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
+
+            // A: output tensor
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
+                in_n_hi_wi_c_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                           make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
+                in_n_hip_wip_c_grid_desc,
+                make_tuple(
+                    make_pass_through_transform(N),
+                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
+                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
+                    make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
+
+            const auto in_gemmktotal_gemmn_grid_desc =
+                transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
+                                            make_tuple(make_merge_transform(make_tuple(Y, X, C)),
+                                                       make_merge_transform(make_tuple(N, Ho, Wo))),
+                                            make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
+                                            make_tuple(Sequence<1>{}, Sequence<0>{}));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+    }
+
+    template <ck::index_t NDim,
+              typename ck::enable_if<NDim == 2 &&
+                                         is_same_v<InLayout, tensor_layout::convolution::NHWGC> &&
+                                         is_same_v<OutLayout, tensor_layout::convolution::NHWGK>,
+                                     bool>::type = false>
+    static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
+    {
+        using namespace ck;
+
+        const index_t Hi = input_spatial_lengths[0];
+        const index_t Wi = input_spatial_lengths[1];
+
+        const index_t Ho = output_spatial_lengths[0];
+        const index_t Wo = output_spatial_lengths[1];
+
+        const index_t Y = filter_spatial_lengths[0];
+        const index_t X = filter_spatial_lengths[1];
+
+        const index_t ConvStrideH = conv_filter_strides[0];
+        const index_t ConvStrideW = conv_filter_strides[1];
+
+        const index_t ConvDilationH = conv_filter_dilations[0];
+        const index_t ConvDilationW = conv_filter_dilations[1];
+
+        const index_t InLeftPadH = input_left_pads[0];
+        const index_t InLeftPadW = input_left_pads[1];
+
+        const index_t InRightPadH = input_right_pads[0];
+        const index_t InRightPadW = input_right_pads[1];
+
+        const index_t GemmKTotal = N * Ho * Wo;
+        const index_t GemmM      = K;
+        const index_t GemmN      = C * X * Y;
+
+        const index_t NStride  = input_strides[1];
+        const index_t HiStride = input_strides[3];
+        const index_t WiStride = input_strides[4];
+        const auto CStride     = input_strides[2];
+
+        const index_t WoStride = output_strides[4];
+        const auto KStride     = Number<1>{};
+
+        const index_t GemmKBatch = batch_k;
+        const index_t GemmK0 =
+            math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
+            K0PerBlock;
+        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // A: output tensor
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Ho * Wo, K), make_tuple(WoStride, KStride));
+
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_gemmktotal_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Hi * Wi, C), make_tuple(WiStride, CStride));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+        else
+        {
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Ho * Wo, K), make_tuple(WoStride, KStride));
+            const auto in_n_hi_wi_c_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N, Hi, Wi, C), make_tuple(NStride, HiStride, WiStride, CStride));
+
+            // A: output tensor
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
+                in_n_hi_wi_c_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                           make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
+                in_n_hip_wip_c_grid_desc,
+                make_tuple(
+                    make_pass_through_transform(N),
+                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
+                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
+                    make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
+
+            const auto in_gemmktotal_gemmn_grid_desc =
+                transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
+                                            make_tuple(make_merge_transform(make_tuple(Y, X, C)),
+                                                       make_merge_transform(make_tuple(N, Ho, Wo))),
+                                            make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
+                                            make_tuple(Sequence<1>{}, Sequence<0>{}));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+    }
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
+    {
+        using namespace ck;
+
+        const index_t Di = input_spatial_lengths[0];
+        const index_t Hi = input_spatial_lengths[1];
+        const index_t Wi = input_spatial_lengths[2];
+
+        const index_t Do = output_spatial_lengths[0];
+        const index_t Ho = output_spatial_lengths[1];
+        const index_t Wo = output_spatial_lengths[2];
+
+        const index_t Z = filter_spatial_lengths[0];
+        const index_t Y = filter_spatial_lengths[1];
+        const index_t X = filter_spatial_lengths[2];
+
+        const index_t ConvStrideD = conv_filter_strides[0];
+        const index_t ConvStrideH = conv_filter_strides[1];
+        const index_t ConvStrideW = conv_filter_strides[2];
+
+        const index_t ConvDilationD = conv_filter_dilations[0];
+        const index_t ConvDilationH = conv_filter_dilations[1];
+        const index_t ConvDilationW = conv_filter_dilations[2];
+
+        const index_t InLeftPadD = input_left_pads[0];
+        const index_t InLeftPadH = input_left_pads[1];
+        const index_t InLeftPadW = input_left_pads[2];
+
+        const index_t InRightPadD = input_right_pads[0];
+        const index_t InRightPadH = input_right_pads[1];
+        const index_t InRightPadW = input_right_pads[2];
+
+        const index_t GemmKTotal = N * Do * Ho * Wo;
+        const index_t GemmM      = K;
+        const index_t GemmN      = C * Z * X * Y;
+
+        const index_t GemmKBatch = batch_k;
+        const index_t GemmK0 =
+            math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
+            K0PerBlock;
+        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // A: output tensor
+            const auto out_gemmktotal_gemmm_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
+
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_gemmktotal_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Di * Hi * Wi, C));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Z * Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+        else
+        {
+            const auto out_gemmktotal_gemmm_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
+            const auto in_n_di_hi_wi_c_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
+
+            // A: output tensor
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_n_dip_hip_wip_c_grid_desc = transform_tensor_descriptor(
+                in_n_di_hi_wi_c_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_pad_transform(Di, InLeftPadD, InRightPadD),
+                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                           make_pass_through_transform(C)),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+            const auto in_n_z_do_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
+                in_n_dip_hip_wip_c_grid_desc,
+                make_tuple(
+                    make_pass_through_transform(N),
+                    make_embed_transform(make_tuple(Z, Do), make_tuple(ConvDilationD, ConvStrideD)),
+                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
+                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
+                    make_pass_through_transform(C)),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                make_tuple(Sequence<0>{},
+                           Sequence<1, 2>{},
+                           Sequence<3, 4>{},
+                           Sequence<5, 6>{},
+                           Sequence<7>{}));
+
+            const auto in_gemmktotal_gemmn_grid_desc = transform_tensor_descriptor(
+                in_n_z_do_y_ho_x_wo_c_grid_desc,
+                make_tuple(make_merge_transform(make_tuple(Z, Y, X, C)),
+                           make_merge_transform(make_tuple(N, Do, Ho, Wo))),
+                make_tuple(Sequence<1, 3, 5, 7>{}, Sequence<0, 2, 4, 6>{}),
+                make_tuple(Sequence<1>{}, Sequence<0>{}));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Z * Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+    } // function end
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
+    static auto GetABCGridDesc()
+    {
+        const ck::index_t dim   = 1;
+        const ck::index_t batch = 1;
+        const std::array<ck::index_t, NDimSpatial> lengths{1};
+        const std::array<ck::index_t, NDimSpatial + 3> strides{1, 1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial> params{1};
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<1>(dim,
+                                                                  dim,
+                                                                  dim,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  strides,
+                                                                  strides,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  batch);
+    }
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
+    static auto GetABCGridDesc()
+    {
+        const ck::index_t dim   = 1;
+        const ck::index_t batch = 1;
+        const std::array<ck::index_t, NDimSpatial> lengths{1, 1};
+        const std::array<ck::index_t, NDimSpatial + 3> strides{1, 1, 1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial> params{1, 1};
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<2>(dim,
+                                                                  dim,
+                                                                  dim,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  strides,
+                                                                  strides,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  batch);
+    }
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    static auto GetABCGridDesc()
+    {
+        const ck::index_t dim   = 1;
+        const ck::index_t batch = 1;
+        const std::array<ck::index_t, NDimSpatial> lengths{1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial + 3> strides{1, 1, 1, 1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial> params{1, 1, 1};
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<3>(dim,
+                                                                  dim,
+                                                                  dim,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  strides,
+                                                                  strides,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  batch);
+    }
+
+    // type convert descs
+    template <typename Desc_M0>
+    static auto PadDescriptor_M0_1d(Desc_M0 desc_m0, index_t gridSize, index_t blockSize)
+    {
+        const auto m0           = desc_m0.GetLength(I0);
+        const index_t loop_step = gridSize * blockSize * 4;
+        const auto pad          = math::integer_least_multiple(m0, loop_step) - m0;
+        const auto desc_m0_pad =
+            transform_tensor_descriptor(desc_m0,
+                                        make_tuple(make_right_pad_transform(m0, pad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+        return desc_m0_pad;
+    }
+
+    template <index_t Dim>
+    static auto MakeDescriptor_M0(const std::array<index_t, Dim>& shape,
+                                  const std::array<index_t, Dim>& stride,
+                                  index_t gridSize,
+                                  index_t blockSize)
+    {
+        auto tupleOfShape  = generate_tuple([&](auto I) { return shape[I]; }, Number<Dim>{});
+        auto tupleOfStride = generate_tuple([&](auto I) { return stride[I]; }, Number<Dim>{});
+
+        // nd desc - [s0, s1, s2, ...]
+        const auto desc = make_naive_tensor_descriptor(tupleOfShape, tupleOfStride);
+
+        // merge nd to 1d desc - [s0 * s1 * ...]
+        if constexpr(Dim > 1)
+        {
+            const auto desc_m0 = transform_tensor_descriptor(
+                desc,
+                make_tuple(make_merge_transform(tupleOfShape)),
+                make_tuple(generate_sequence_v2([&](auto I) { return I; }, Number<Dim>{})),
+                make_tuple(Sequence<0>{}));
+
+            return PadDescriptor_M0_1d(desc_m0, gridSize, blockSize);
+        }
+        else
+            return PadDescriptor_M0_1d(desc, gridSize, blockSize);
+    }
+
+    using GridDesc_M0 = decltype(MakeDescriptor_M0<1>({1}, {1}, 1, 1));
+
+    using ABCGridDescs = decltype(GetABCGridDesc<NDimSpatial>());
+
+    using AGridDesc_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>;
+    using BGridDesc_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>;
+    using CGridDesc_M_N     = remove_cvref_t<decltype(ABCGridDescs{}[I2])>;
+
+    using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight<
+        BlockSize,
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        CDataType,
+        InMemoryDataOperationEnum::AtomicAdd,
+        AGridDesc_K0_M_K1,
+        BGridDesc_K0_N_K1,
+        CGridDesc_M_N,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerXdl,
+        NPerXdl,
+        K1,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        ABlockLdsM1PerBlock,
+        ABlockLdsM0PerBlock,
+        ABlockLdsM1Padding,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        BBlockLdsN1PerBlock,
+        BBlockLdsN0PerBlock,
+        BBlockLdsN1Padding,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        CBlockTransferScalarPerVector_NWaveNPerXdl,
+        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        true,
+        true>;
+
+    // Argument
+    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}));
+
+    using Block2CTileMap =
+        decltype(GridwiseGemm::MakeCBlockClusterAdaptor(CGridDesc_M_N{}, 1, 1, 1));
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const InDataType* p_in_grid,
+                 WeiDataType* p_wei_grid,
+                 const OutDataType* p_out_grid,
+                 const ck::index_t G,
+                 const ck::index_t N,
+                 const ck::index_t K,
+                 const ck::index_t C,
+                 const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                 const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                 const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+                 const ck::index_t M01,
+                 const ck::index_t N01,
+                 InElementwiseOperation in_element_op,
+                 WeiElementwiseOperation wei_element_op,
+                 OutElementwiseOperation out_element_op,
+                 ck::index_t split_k)
+            : p_a_grid_{p_out_grid},
+              p_b_grid_{p_in_grid},
+              p_c_grid_{p_wei_grid},
+              a_grid_desc_kbatch_k0_m_k1_{},
+              b_grid_desc_kbatch_k0_n_k1_{},
+              c_grid_desc_m_n_{},
+              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              block_2_ctile_map_{},
+              compute_ptr_offset_of_batch_{},
+              M01_{M01},
+              N01_{N01},
+              a_element_op_{out_element_op},
+              b_element_op_{in_element_op},
+              c_element_op_{wei_element_op},
+              Conv_G_{G},
+              Conv_N_{N},
+              Conv_K_{K},
+              Conv_C_{C},
+              output_spatial_lengths_{output_spatial_lengths},
+              filter_spatial_lengths_{filter_spatial_lengths},
+              conv_filter_strides_{conv_filter_strides},
+              input_left_pads_{input_left_pads},
+              input_right_pads_{input_right_pads},
+              k_batch_{split_k}
+        {
+            const auto descs =
+                DeviceOp::MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<NDimSpatial>(
+                    N,
+                    K,
+                    C,
+                    input_spatial_lengths,
+                    filter_spatial_lengths,
+                    output_spatial_lengths,
+                    input_strides,
+                    output_strides,
+                    conv_filter_strides,
+                    conv_filter_dilations,
+                    input_left_pads,
+                    input_right_pads,
+                    k_batch_);
+
+            a_grid_desc_kbatch_k0_m_k1_ = descs[I0];
+            b_grid_desc_kbatch_k0_n_k1_ = descs[I1];
+            c_grid_desc_m_n_            = descs[I2];
+
+            block_2_ctile_map_ =
+                GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
+
+            // A/B/C Batch Stride
+            compute_ptr_offset_of_batch_.BatchStrideA_ = output_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideB_ = input_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideC_ =
+                K * C *
+                std::accumulate(begin(filter_spatial_lengths),
+                                end(filter_spatial_lengths),
+                                index_t{1},
+                                std::multiplies<>{});
+
+            if(GridwiseGemm::CheckValidity(a_grid_desc_kbatch_k0_m_k1_,
+                                           b_grid_desc_kbatch_k0_n_k1_,
+                                           c_grid_desc_m_n_,
+                                           block_2_ctile_map_))
+            {
+                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n_);
+            }
+        }
+
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        CDataType* p_c_grid_;
+        AGridDesc_K0_M_K1 a_grid_desc_kbatch_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_kbatch_k0_n_k1_;
+        CGridDesc_M_N c_grid_desc_m_n_;
+        CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock c_grid_desc_mblock_mperblock_nblock_nperblock_;
+
+        Block2CTileMap block_2_ctile_map_;
+
+        // for computing batch offset
+        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
+
+        index_t M01_;
+        index_t N01_;
+
+        InElementwiseOperation a_element_op_;
+        OutElementwiseOperation b_element_op_;
+        WeiElementwiseOperation c_element_op_;
+
+        // for checking IsSupportedArgument()
+        const index_t Conv_G_;
+        const index_t Conv_N_;
+        const index_t Conv_K_;
+        const index_t Conv_C_;
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides_;
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads_;
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads_;
+        const index_t k_batch_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        void ShowInfo(const Argument& arg)
+        {
+            std::cout << "arg.a_grid_desc_kbatch_k0_m_k1_{"
+                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I0) << ", "
+                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1) << ", "
+                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I2) << ", "
+                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I3) << "}" << std::endl;
+
+            std::cout << "arg.b_grid_desc_kbatch_k0_n_k1_{"
+                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I0) << ", "
+                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I1) << ", "
+                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I2) << ", "
+                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I3) << "}" << std::endl;
+
+            std::cout << "arg.c_grid_desc_m_n_{" << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
+                      << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+        }
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
+                                            arg.b_grid_desc_kbatch_k0_n_k1_,
+                                            arg.c_grid_desc_m_n_,
+                                            arg.block_2_ctile_map_))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v3r1 has invalid setting");
+            }
+
+            const index_t grid_size =
+                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_) * arg.Conv_G_;
+
+            const auto K0 = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1);
+
+            const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel = kernel_batched_gemm_xdlops_bwd_weight<
+                    GridwiseGemm,
+                    ADataType, // TODO: distiguish A/B datatype
+                    CDataType,
+                    OutElementwiseOperation,
+                    InElementwiseOperation,
+                    WeiElementwiseOperation,
+                    remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
+                    remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
+                    remove_reference_t<DeviceOp::CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
+                    remove_reference_t<DeviceOp::Block2CTileMap>,
+                    ComputePtrOffsetOfStridedBatch,
+                    has_main_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_a_grid_,
+                                              arg.p_b_grid_,
+                                              arg.p_c_grid_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.c_element_op_,
+                                              arg.Conv_G_,
+                                              arg.a_grid_desc_kbatch_k0_m_k1_,
+                                              arg.b_grid_desc_kbatch_k0_n_k1_,
+                                              arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.block_2_ctile_map_,
+                                              arg.compute_ptr_offset_of_batch_);
+            };
+
+            if(has_main_k0_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if constexpr(NDimSpatial == 1)
+        {
+            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNWC> &&
+                           is_same_v<WeiLayout, tensor_layout::convolution::GKXC> &&
+                           is_same_v<OutLayout, tensor_layout::convolution::GNWK>))
+            {
+                return false;
+            }
+        }
+        else if constexpr(NDimSpatial == 2)
+        {
+            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
+                           is_same_v<WeiLayout, tensor_layout::convolution::GKYXC> &&
+                           is_same_v<OutLayout,
+                                     tensor_layout::convolution::
+                                         GNHWK>)&&!(is_same_v<InLayout,
+                                                              tensor_layout::convolution::NHWGC> &&
+                                                    is_same_v<WeiLayout,
+                                                              tensor_layout::convolution::GKYXC> &&
+                                                    is_same_v<OutLayout,
+                                                              tensor_layout::convolution::NHWGK>))
+            {
+                return false;
+            }
+        }
+        else if constexpr(NDimSpatial == 3)
+        {
+            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNDHWC> &&
+                           is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+                           is_same_v<OutLayout, tensor_layout::convolution::GNDHWK>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // check if it's 1x1, stride=1 pad = 0 conv
+            for(int i = 0; i < NDimSpatial; i++)
+            {
+                if(!(arg.filter_spatial_lengths_[i] == 1 && arg.conv_filter_strides_[i] == 1 &&
+                     arg.input_left_pads_[i] == 0 && arg.input_right_pads_[i] == 0))
+                {
+                    return false;
+                }
+            }
+        }
+
+        // vector load A/B matrix from global memory
+        if(!(ABlockTransferSrcVectorDim == 2 && BBlockTransferSrcVectorDim == 2 &&
+             arg.Conv_K_ % ABlockTransferSrcScalarPerVector == 0 &&
+             arg.Conv_C_ % BBlockTransferSrcScalarPerVector == 0))
+        {
+            return false;
+        }
+
+        // vector store C matrix into global memory
+        if(!(arg.Conv_C_ % CBlockTransferScalarPerVector_NWaveNPerXdl == 0))
+        {
+            return false;
+        }
+
+        // Gridwise GEMM size
+        return GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
+                                           arg.b_grid_desc_kbatch_k0_n_k1_,
+                                           arg.c_grid_desc_m_n_,
+                                           arg.block_2_ctile_map_);
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const InDataType* p_in_grid,
+                             WeiDataType* p_wei_grid,
+                             const OutDataType* p_out_grid,
+                             const ck::index_t G,
+                             const ck::index_t N,
+                             const ck::index_t K,
+                             const ck::index_t C,
+                             const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                             const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                             const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                             const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+                             InElementwiseOperation in_element_op,
+                             WeiElementwiseOperation wei_element_op,
+                             OutElementwiseOperation out_element_op,
+                             const ck::index_t split_k)
+    {
+        return Argument{p_in_grid,
+                        p_wei_grid,
+                        p_out_grid,
+                        G,
+                        N,
+                        K,
+                        C,
+                        input_spatial_lengths,
+                        filter_spatial_lengths,
+                        output_spatial_lengths,
+                        input_strides,
+                        output_strides,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads,
+                        1,
+                        1,
+                        in_element_op,
+                        wei_element_op,
+                        out_element_op,
+                        split_k};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in_grid,
+                        void* p_wei_grid,
+                        const void* p_out_grid,
+                        const ck::index_t G,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t C,
+                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+                        InElementwiseOperation in_element_op,
+                        WeiElementwiseOperation wei_element_op,
+                        OutElementwiseOperation out_element_op,
+                        const ck::index_t split_k) override
+    {
+        return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_grid),
+                                          static_cast<WeiDataType*>(p_wei_grid),
+                                          static_cast<const OutDataType*>(p_out_grid),
+                                          G,
+                                          N,
+                                          K,
+                                          C,
+                                          input_spatial_lengths,
+                                          filter_spatial_lengths,
+                                          output_spatial_lengths,
+                                          input_strides,
+                                          output_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          1,
+                                          1,
+                                          in_element_op,
+                                          wei_element_op,
+                                          out_element_op,
+                                          split_k);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedConvBwdWeight_Xdl_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << getConvBackwardWeightSpecializationString(ConvBackwardWeightSpecialization) << ", "
+            << K1 << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
similarity index 98%
rename from include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index 079135e5b2fb59ed6e49ab4ed8d1e7a348cc0a24..ea958a4eb5972a72430c2076bd3b7def2999a3ee 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -134,7 +134,9 @@ __global__ void
             const Block2CTileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) ||           \
+    defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx940__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -314,9 +316,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
         const auto in_gemmm_gemmk_desc =
             matrix_padder.PadADescriptor_M_K(in_gemmmraw_gemmkraw_desc);
 
-        const auto M = in_gemmm_gemmk_desc.GetLength(I0);
-        const auto K = in_gemmm_gemmk_desc.GetLength(I1);
-
+        const auto M   = in_gemmm_gemmk_desc.GetLength(I0);
+        const auto K   = in_gemmm_gemmk_desc.GetLength(I1);
         const auto AK0 = K / K1;
 
         return transform_tensor_descriptor(
@@ -709,7 +710,11 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
         namespace ctc = tensor_layout::convolution;
 
         // check device
-        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030"))
+        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+             ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx908" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx1100" ||
+             ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102" ||
+             ck::get_device_name() == "gfx941" || ck::get_device_name() == "gfx942"))
         {
             return false;
         }
@@ -834,6 +839,7 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
         {
             return false;
         }
+
         // check Gridwise GEMM
         return GridwiseGemm::CheckValidity(
             arg.a_grid_desc_ak0_m_ak1_, arg.b_grid_desc_bk0_n_bk1_, arg.e_grid_desc_m_n_);
@@ -946,7 +952,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
             << MPerBlock << ", "
             << NPerBlock << ", "
             << K0PerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << K1
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
similarity index 97%
rename from include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
index 47c821171beecd3f3745866885c067c27caa856d..9080936589531fb7c8afbcc84dabcace97bfc812 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -106,7 +106,8 @@ __global__ void
             const Block2CTileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) || \
+    defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -600,7 +601,9 @@ struct DeviceGroupedConvFwdDl_NHWC_KYXC_NHWK : public DeviceGroupedConvFwd<NDimS
         namespace ctc = tensor_layout::convolution;
 
         // check device
-        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030"))
+        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+             ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+             ck::get_device_name() == "gfx1102"))
         {
             return false;
         }
@@ -824,7 +827,19 @@ struct DeviceGroupedConvFwdDl_NHWC_KYXC_NHWK : public DeviceGroupedConvFwd<NDimS
             << MPerBlock << ", "
             << NPerBlock << ", "
             << K0PerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << CBlockTransferScalarPerVector_NWaveNPerXdl
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp
index 03185d5b1d29186f2b2814257eacfe90da6c5e02..face627e1fb85d0621823803eabc5409b0e73d61 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
index 8b54ee493c417983b01da6d27b19e5a557de5431..af8e0e02794def75284258a09609baefc688ebe9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -155,7 +155,8 @@ __global__ void
             const Block2ETileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -810,7 +811,8 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
                 return false;
             }
         }
-        else if(get_device_name() == "gfx90a")
+        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940" ||
+                get_device_name() == "gfx941" || get_device_name() == "gfx942")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
@@ -1092,7 +1094,15 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
             << MPerBlock << ", "
             << NPerBlock << ", "
             << KPerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..526108b87442b4077bc9723820edcd09b7fd76f3
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
@@ -0,0 +1,854 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <functional>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/io.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+namespace {
+
+template <index_t NumDTensor>
+struct ComputePtrOffsetOfStridedBatch
+{
+    ComputePtrOffsetOfStridedBatch() = default;
+
+    ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
+                                   index_t BatchStrideB,
+                                   Array<ck::index_t, NumDTensor> BatchStrideDs,
+                                   index_t BatchStrideE)
+        : BatchStrideA_(BatchStrideA),
+          BatchStrideB_(BatchStrideB),
+          BatchStrideDs_(BatchStrideDs),
+          BatchStrideE_(BatchStrideE)
+    {
+    }
+
+    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideA_);
+    }
+
+    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideB_);
+    }
+
+    __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
+    {
+        Array<long_index_t, NumDTensor> ds_offset;
+        static_for<0, NumDTensor, 1>{}(
+            [&](auto i) { ds_offset(i) = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]); });
+        return ds_offset;
+    }
+
+    __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideE_);
+    }
+
+    index_t BatchStrideA_;
+    index_t BatchStrideB_;
+    Array<ck::index_t, NumDTensor> BatchStrideDs_;
+    index_t BatchStrideE_;
+};
+
+} // namespace
+
+//
+// @brief      Device Convolution operation.
+//
+// Supports:
+//  @li         Forward convolution with up to 3 spatial dimentions
+//  @li         Input tensor in GNWC data format
+//  @li         Weight tensor in GKXC data format
+//  @li         Output tensor in GNWK data format
+//
+// 1D:
+// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
+// 2D:
+// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+// 3D:
+// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
+// Assume:
+//  AK1 == BK1
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          ConvolutionForwardSpecialization ConvForwardSpecialization,
+          GemmSpecialization GemmSpec,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t K0PerBlock,
+          ck::index_t K1,
+          ck::index_t MPerWMMA,
+          ck::index_t NPerWMMA,
+          ck::index_t MRepeat,
+          ck::index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+          index_t NumGemmKPrefetchStage   = 1,
+          LoopScheduler LoopSched         = make_default_loop_scheduler(),
+          ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
+struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
+    : public DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                           ALayout,
+                                           BLayout,
+                                           DsLayout,
+                                           ELayout,
+                                           ADataType,
+                                           BDataType,
+                                           DsDataType,
+                                           EDataType,
+                                           AElementwiseOperation,
+                                           BElementwiseOperation,
+                                           CDEElementwiseOperation>
+{
+    using DeviceOp = DeviceGroupedConvFwdMultipleD_Wmma_CShuffle;
+
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0           = Number<0>{};
+    static constexpr auto I1           = Number<1>{};
+    static constexpr auto I2           = Number<2>{};
+    static constexpr auto I3           = Number<3>{};
+    static constexpr index_t KPerBlock = K0PerBlock * K1;
+
+    static constexpr auto conv_to_gemm_transformer =
+        TransformConvFwdToGemm<NDimSpatial, ConvForwardSpecialization>{};
+
+    static constexpr auto matrix_padder =
+        MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock};
+
+    template <typename ALay>
+    static auto
+    MakeAGridDescriptor_M_K(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
+                            const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                            const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
+                            const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                            const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths,
+                            const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+                            const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                            const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                            const std::array<index_t, NDimSpatial>& input_left_pads,
+                            const std::array<index_t, NDimSpatial>& input_right_pads)
+    {
+        const auto in_gemmmraw_gemmkraw_desc =
+            conv_to_gemm_transformer.template MakeADescriptor_M_K<ALay>(a_g_n_c_wis_lengths,
+                                                                        a_g_n_c_wis_strides,
+                                                                        b_g_k_c_xs_lengths,
+                                                                        b_g_k_c_xs_strides,
+                                                                        e_g_n_k_wos_lengths,
+                                                                        e_g_n_k_wos_strides,
+                                                                        conv_filter_strides,
+                                                                        conv_filter_dilations,
+                                                                        input_left_pads,
+                                                                        input_right_pads);
+
+        const auto in_gemmm_gemmk_desc =
+            matrix_padder.PadADescriptor_M_K(in_gemmmraw_gemmkraw_desc);
+
+        return in_gemmm_gemmk_desc;
+    }
+
+    template <typename BLay>
+    static auto
+    MakeBGridDescriptor_N_K(const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
+                            const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides)
+    {
+        const auto wei_gemmnraw_gemmkraw_desc =
+            conv_to_gemm_transformer.template MakeBDescriptor_N_K<BLay>(b_g_k_c_xs_lengths,
+                                                                        b_g_k_c_xs_strides);
+
+        const auto wei_gemmn_gemmk_desc =
+            matrix_padder.PadBDescriptor_N_K(wei_gemmnraw_gemmkraw_desc);
+
+        return wei_gemmn_gemmk_desc;
+    }
+
+    template <typename ELay>
+    static auto
+    MakeEGridDescriptor_M_N(const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths,
+                            const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides)
+    {
+        const auto out_gemmmraw_gemmnraw_desc =
+            conv_to_gemm_transformer.template MakeCDescriptor_M_N<ELay>(e_g_n_k_wos_lengths,
+                                                                        e_g_n_k_wos_strides);
+
+        const auto out_gemmm_gemmn_desc =
+            matrix_padder.PadCDescriptor_M_N(out_gemmmraw_gemmnraw_desc);
+
+        return out_gemmm_gemmn_desc;
+    }
+
+    static auto MakeDsGridDescriptor_M_N(
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>& ds_g_n_k_wos_lengths,
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>& ds_g_n_k_wos_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return DeviceOp::MakeEGridDescriptor_M_N<DLayout>(ds_g_n_k_wos_lengths[i],
+                                                                  ds_g_n_k_wos_strides[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // desc for problem definition
+    using AGridDesc_M_K  = remove_cvref_t<decltype(
+        MakeAGridDescriptor_M_K<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using BGridDesc_N_K  = remove_cvref_t<decltype(MakeBGridDescriptor_N_K<BLayout>({}, {}))>;
+    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
+    using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
+
+    // A desc for source in blockwise copy
+    template <typename AGridDesc_M_K>
+    __host__ __device__ static constexpr auto
+    MakeAGridDescriptor_AK0_M_AK1(const AGridDesc_M_K& a_grid_desc_m_k)
+    {
+        const auto M = a_grid_desc_m_k.GetLength(I0);
+        const auto K = a_grid_desc_m_k.GetLength(I1);
+
+        const auto AK1 = K1;
+        const auto AK0 = K / AK1;
+
+        return transform_tensor_descriptor(a_grid_desc_m_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
+                                                      make_pass_through_transform(M)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    // B desc for source in blockwise copy
+    template <typename BGridDesc_N_K>
+    __host__ __device__ static constexpr auto
+    MakeBGridDescriptor_BK0_N_BK1(const BGridDesc_N_K& b_grid_desc_n_k)
+    {
+        const auto N = b_grid_desc_n_k.GetLength(I0);
+        const auto K = b_grid_desc_n_k.GetLength(I1);
+
+        const auto BK1 = K1;
+        const auto BK0 = K / BK1;
+
+        return transform_tensor_descriptor(b_grid_desc_n_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                                                      make_pass_through_transform(N)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    using AGridDesc_AK0_M_AK1 = decltype(DeviceOp::MakeAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}));
+    using BGridDesc_BK0_N_BK1 = decltype(DeviceOp::MakeBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}));
+
+    // GridwiseOp
+    using GridwiseOp = GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle<
+        // DataType Family
+        ADataType,
+        BDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        // InMemory Data Descriptor
+        AGridDesc_AK0_M_AK1,
+        BGridDesc_BK0_N_BK1,
+        DsGridDesc_M_N,
+        EGridDesc_M_N,
+        // ElementwiseOp Family
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        // Tiling Family
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerWMMA,
+        NPerWMMA,
+        K1,
+        MRepeat,
+        NRepeat,
+        // ThreadCluster Family
+        BlockSize,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+        NumGemmKPrefetchStage,
+        LoopSched,
+        PipelineVer>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a,
+                 const void* p_b,
+                 const std::array<const void*, NumDTensor>& p_ds,
+                 void* p_e,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                 const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+                     ds_g_n_k_wos_lengths,
+                 const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+                     ds_g_n_k_wos_strides,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<index_t, NDimSpatial>& input_left_pads,
+                 const std::array<index_t, NDimSpatial>& input_right_pads,
+                 index_t M01,
+                 index_t N01,
+                 const AElementwiseOperation& a_element_op,
+                 const BElementwiseOperation& b_element_op,
+                 const CDEElementwiseOperation& cde_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a)},
+              p_b_grid_{static_cast<const BDataType*>(p_b)},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e)},
+              num_group_{a_g_n_c_wis_lengths[0]},
+              a_grid_desc_m_k_{DeviceOp::MakeAGridDescriptor_M_K<ALayout>(a_g_n_c_wis_lengths,
+                                                                          a_g_n_c_wis_strides,
+                                                                          b_g_k_c_xs_lengths,
+                                                                          b_g_k_c_xs_strides,
+                                                                          e_g_n_k_wos_lengths,
+                                                                          e_g_n_k_wos_strides,
+                                                                          conv_filter_strides,
+                                                                          conv_filter_dilations,
+                                                                          input_left_pads,
+                                                                          input_right_pads)},
+              b_grid_desc_n_k_{DeviceOp::MakeBGridDescriptor_N_K<BLayout>(b_g_k_c_xs_lengths,
+                                                                          b_g_k_c_xs_strides)},
+              ds_grid_desc_m_n_{},
+              e_grid_desc_m_n_{DeviceOp::MakeEGridDescriptor_M_N<ELayout>(e_g_n_k_wos_lengths,
+                                                                          e_g_n_k_wos_strides)},
+              a_grid_desc_ak0_m_ak1_{DeviceOp::MakeAGridDescriptor_AK0_M_AK1(a_grid_desc_m_k_)},
+              b_grid_desc_bk0_n_bk1_{DeviceOp::MakeBGridDescriptor_BK0_N_BK1(b_grid_desc_n_k_)},
+              ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              e_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              block_2_etile_map_{GridwiseOp::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_, M01, N01)},
+              compute_ptr_offset_of_batch_{},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op},
+              a_g_n_c_wis_lengths_{a_g_n_c_wis_lengths},
+              a_g_n_c_wis_strides_{a_g_n_c_wis_strides},
+              b_g_k_c_xs_lengths_{b_g_k_c_xs_lengths},
+              b_g_k_c_xs_strides_{b_g_k_c_xs_strides},
+              ds_g_n_k_wos_lengths_{ds_g_n_k_wos_lengths},
+              ds_g_n_k_wos_strides_{ds_g_n_k_wos_strides},
+              e_g_n_k_wos_lengths_{e_g_n_k_wos_lengths},
+              e_g_n_k_wos_strides_{e_g_n_k_wos_strides},
+              conv_filter_strides_{conv_filter_strides},
+              conv_filter_dilations_{conv_filter_dilations},
+              input_left_pads_{input_left_pads},
+              input_right_pads_{input_right_pads}
+        {
+            // A/B/E Batch Stride
+            compute_ptr_offset_of_batch_.BatchStrideA_ = a_g_n_c_wis_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideB_ = b_g_k_c_xs_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideE_ = e_g_n_k_wos_strides[0];
+
+            // populate pointer, batch stride, desc for Ds
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                // using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds[i]);
+
+                // D batch stride
+                compute_ptr_offset_of_batch_.BatchStrideDs_(i) = ds_g_n_k_wos_strides[i][0];
+            });
+
+            // D desc
+            ds_grid_desc_m_n_ =
+                DeviceOp::MakeDsGridDescriptor_M_N(ds_g_n_k_wos_lengths, ds_g_n_k_wos_strides);
+
+            // populate desc for Ds/E
+            e_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                GridwiseOp::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(e_grid_desc_m_n_);
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                GridwiseOp::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                    ds_grid_desc_m_n_);
+        }
+
+        void Print() const
+        {
+            std::cout << "A[M, K]: " << a_grid_desc_m_k_ << std::endl;
+            std::cout << "B[N, K]: " << b_grid_desc_n_k_ << std::endl;
+            static_for<0, NumDTensor, 1>{}(
+                [&](auto i) { std::cout << "Ds[M, N]: " << ds_grid_desc_m_n_[i] << std::endl; });
+            std::cout << "E[M, N]: " << e_grid_desc_m_n_ << std::endl;
+        }
+
+        //  private:
+        // pointers
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseOp::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        // tensor descriptors for problem definiton
+        index_t num_group_;
+        AGridDesc_M_K a_grid_desc_m_k_;
+        BGridDesc_N_K b_grid_desc_n_k_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        // tensor descriptors for block/thread-wise copy
+        AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
+        BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
+        typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
+        typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+            e_grid_desc_mblock_mperblock_nblock_nperblock_;
+
+        // block-to-e-tile map
+        typename GridwiseOp::DefaultBlock2CTileMap block_2_etile_map_;
+
+        // for computing batch offset
+        ComputePtrOffsetOfStridedBatch<NumDTensor> compute_ptr_offset_of_batch_;
+
+        // element-wise op
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+
+        // for checking IsSupportedArgument()
+        std::array<index_t, NDimSpatial + 3> a_g_n_c_wis_lengths_;
+        std::array<index_t, NDimSpatial + 3> a_g_n_c_wis_strides_;
+        std::array<index_t, NDimSpatial + 3> b_g_k_c_xs_lengths_;
+        std::array<index_t, NDimSpatial + 3> b_g_k_c_xs_strides_;
+        std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor> ds_g_n_k_wos_lengths_;
+        std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor> ds_g_n_k_wos_strides_;
+        std::array<index_t, NDimSpatial + 3> e_g_n_k_wos_lengths_;
+        std::array<index_t, NDimSpatial + 3> e_g_n_k_wos_strides_;
+        std::array<index_t, NDimSpatial> conv_filter_strides_;
+        std::array<index_t, NDimSpatial> conv_filter_dilations_;
+        std::array<index_t, NDimSpatial> input_left_pads_;
+        std::array<index_t, NDimSpatial> input_right_pads_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(stream_config.log_level_ > 0)
+            {
+                arg.Print();
+            }
+
+            const index_t grid_size =
+                arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_) * arg.num_group_;
+
+            const auto K =
+                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel = kernel_grouped_conv_fwd_multiple_d_wmma_cshuffle<
+                    GridwiseOp,
+                    ADataType,
+                    BDataType,
+                    typename GridwiseOp::DsGridPointer,
+                    EDataType,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CDEElementwiseOperation,
+                    DeviceOp::AGridDesc_AK0_M_AK1,
+                    DeviceOp::BGridDesc_BK0_N_BK1,
+                    typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                    typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                    remove_reference_t<typename GridwiseOp::DefaultBlock2CTileMap>,
+                    ComputePtrOffsetOfStridedBatch<NumDTensor>,
+                    has_main_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_a_grid_,
+                                              arg.p_b_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.a_g_n_c_wis_lengths_[0], // Group count
+                                              arg.a_grid_desc_ak0_m_ak1_,
+                                              arg.b_grid_desc_bk0_n_bk1_,
+                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.block_2_etile_map_,
+                                              arg.compute_ptr_offset_of_batch_);
+            };
+
+            if(GridwiseOp::CalculateHasMainKBlockLoop(K))
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        namespace ctc = tensor_layout::convolution;
+
+        // check device
+        if(get_device_name() == "gfx1100" || get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
+        {
+            if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // check ConvolutionForwardSpecialization
+        if constexpr(ConvForwardSpecialization ==
+                     ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
+        {
+            // check if it's 1x1, stride=1 conv
+            for(index_t i = 0; i < NDimSpatial; ++i)
+            {
+                const index_t X          = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t ConvStride = arg.conv_filter_strides_[i];
+                const index_t LeftPad    = arg.input_left_pads_[i];
+                const index_t RightPad   = arg.input_right_pads_[i];
+
+                if(!(X == 1 && ConvStride == 1 && LeftPad == 0 && RightPad == 0))
+                {
+                    return false;
+                }
+            }
+        }
+        else if constexpr(ConvForwardSpecialization ==
+                          ConvolutionForwardSpecialization::Filter1x1Pad0)
+        {
+            // check if it's 1x1 conv
+            for(index_t i = 0; i < NDimSpatial; ++i)
+            {
+                const index_t X        = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t LeftPad  = arg.input_left_pads_[i];
+                const index_t RightPad = arg.input_right_pads_[i];
+
+                if(!(X == 1 && LeftPad == 0 && RightPad == 0))
+                {
+                    return false;
+                }
+            }
+        }
+
+        // check vector access of A
+        // FIXME: layout
+        if constexpr(is_same_v<ALayout, ctc::G_NW_C> || is_same_v<ALayout, ctc::G_NHW_C> ||
+                     is_same_v<ALayout, ctc::G_NDHW_C> || is_same_v<ALayout, ctc::GNWC> ||
+                     is_same_v<ALayout, ctc::GNHWC> || is_same_v<ALayout, ctc::GNDHWC> ||
+                     is_same_v<ALayout, ctc::NWGC> || is_same_v<ALayout, ctc::NHWGC> ||
+                     is_same_v<ALayout, ctc::NDHWGC>)
+        {
+            const index_t C = arg.a_g_n_c_wis_lengths_[2];
+
+            if(!(ABlockTransferSrcVectorDim == 2 && C % ABlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // check vector access of B
+        // FIXME: layout
+        if constexpr(is_same_v<BLayout, ctc::G_K_X_C> || is_same_v<BLayout, ctc::G_K_YX_C> ||
+                     is_same_v<BLayout, ctc::G_K_ZYX_C> || is_same_v<BLayout, ctc::GKXC> ||
+                     is_same_v<BLayout, ctc::GKYXC> || is_same_v<BLayout, ctc::GKZYXC> ||
+                     is_same_v<BLayout, ctc::KXGC> || is_same_v<BLayout, ctc::KYXGC> ||
+                     is_same_v<BLayout, ctc::KZYXGC>)
+
+        {
+            const index_t C = arg.b_g_k_c_xs_lengths_[2];
+
+            if(!(BBlockTransferSrcVectorDim == 2 && C % BBlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        //  check vector access of Ds
+        bool valid = true;
+
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+            // FIXME: layout
+            if constexpr(is_same_v<DLayout, ctc::G_NW_K> || is_same_v<DLayout, ctc::G_NHW_K> ||
+                         is_same_v<DLayout, ctc::G_NDHW_K> || is_same_v<DLayout, ctc::GNWK> ||
+                         is_same_v<DLayout, ctc::GNHWK> || is_same_v<DLayout, ctc::GNDHWK> ||
+                         is_same_v<DLayout, ctc::NWGK> || is_same_v<DLayout, ctc::NHWGK> ||
+                         is_same_v<DLayout, ctc::NDHWGK> || is_same_v<DLayout, ctc::GK> ||
+                         is_same_v<DLayout, ctc::G_K>)
+            {
+                const index_t K = arg.ds_g_n_k_wos_lengths_[i][2];
+
+                if(!(K % CDEShuffleBlockTransferScalarPerVector_NPerBlock == 0))
+                {
+                    valid = false;
+                }
+            }
+            else
+            {
+                valid = false;
+            }
+        });
+
+        if(!valid)
+        {
+            return false;
+        }
+
+        // check vector access of E
+        if constexpr(is_same_v<ELayout, ctc::G_NW_K> || is_same_v<ELayout, ctc::G_NHW_K> ||
+                     is_same_v<ELayout, ctc::G_NDHW_K> || is_same_v<ELayout, ctc::GNWK> ||
+                     is_same_v<ELayout, ctc::GNHWK> || is_same_v<ELayout, ctc::GNDHWK> ||
+                     is_same_v<ELayout, ctc::NWGK> || is_same_v<ELayout, ctc::NHWGK> ||
+                     is_same_v<ELayout, ctc::NDHWGK>)
+        {
+            const index_t K = arg.e_g_n_k_wos_lengths_[2];
+
+            if(!(K % CDEShuffleBlockTransferScalarPerVector_NPerBlock == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // check Gridwise GEMM
+        return GridwiseOp::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
+                                         arg.b_grid_desc_bk0_n_bk1_,
+                                         arg.ds_grid_desc_m_n_,
+                                         arg.e_grid_desc_m_n_,
+                                         arg.block_2_etile_map_);
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(
+        const void* p_a,
+        const void* p_b,
+        const std::array<const void*, NumDTensor>& p_ds,
+        void* p_e,
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>& ds_g_n_k_wos_lengths,
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>& ds_g_n_k_wos_strides,
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths,
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<index_t, NDimSpatial>& input_left_pads,
+        const std::array<index_t, NDimSpatial>& input_right_pads,
+        const AElementwiseOperation& a_element_op,
+        const BElementwiseOperation& b_element_op,
+        const CDEElementwiseOperation& cde_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_e,
+                        a_g_n_c_wis_lengths,
+                        a_g_n_c_wis_strides,
+                        b_g_k_c_xs_lengths,
+                        b_g_k_c_xs_strides,
+                        ds_g_n_k_wos_lengths,
+                        ds_g_n_k_wos_strides,
+                        e_g_n_k_wos_lengths,
+                        e_g_n_k_wos_strides,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads,
+                        1,
+                        1,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(
+        const void* p_a,
+        const void* p_b,
+        const std::array<const void*, NumDTensor>& p_ds,
+        void* p_e,
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>& ds_g_n_k_wos_lengths,
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>& ds_g_n_k_wos_strides,
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths,
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<index_t, NDimSpatial>& input_left_pads,
+        const std::array<index_t, NDimSpatial>& input_right_pads,
+        const AElementwiseOperation& a_element_op,
+        const BElementwiseOperation& b_element_op,
+        const CDEElementwiseOperation& cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          a_g_n_c_wis_lengths,
+                                          a_g_n_c_wis_strides,
+                                          b_g_k_c_xs_lengths,
+                                          b_g_k_c_xs_strides,
+                                          ds_g_n_k_wos_lengths,
+                                          ds_g_n_k_wos_strides,
+                                          e_g_n_k_wos_lengths,
+                                          e_g_n_k_wos_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          1,
+                                          1,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedConvFwdMultipleD_Wmma_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << K1 << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index bb7a2f8c0a78963fd25380c50bd32615581eb8fe..a9294f0344efa5f3a3fc522f3f3e6544def80027 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -135,7 +135,8 @@ __global__ void
             const Block2ETileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -684,7 +685,8 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
                 return false;
             }
         }
-        else if(get_device_name() == "gfx90a")
+        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940" ||
+                get_device_name() == "gfx941" || get_device_name() == "gfx942")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
@@ -939,7 +941,15 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
             << MPerBlock << ", "
             << NPerBlock << ", "
             << KPerBlock << ", "
-            << getConvForwardSpecializationString(ConvForwardSpecialization)
+            << getConvForwardSpecializationString(ConvForwardSpecialization) << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0190b3cee6b44af7656c34be866c9ab0855f7b0d
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
@@ -0,0 +1,766 @@
+#pragma once
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename GridwiseGemm,
+          typename GemmDesc,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          bool HasMainKBlockLoop,
+          bool HasDoubleTailKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_grouped_gemm_multiple_d_dl(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
+                                          const index_t group_count,
+                                          const AElementwiseOperation a_element_op,
+                                          const BElementwiseOperation b_element_op,
+                                          const CDEElementwiseOperation cde_element_op)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||              \
+    defined(__gfx90a__) || defined(__gfx1030__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    const index_t block_id = get_block_1d_id();
+
+    const auto gemm_desc_ptr =
+        reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
+
+    index_t left     = 0;
+    index_t right    = group_count;
+    index_t group_id = index_t((left + right) / 2);
+    while((!(block_id >= gemm_desc_ptr[group_id].BlockStart_ &&
+             block_id < gemm_desc_ptr[group_id].BlockEnd_)) &&
+          left <= right)
+    {
+        if(block_id < gemm_desc_ptr[group_id].BlockStart_)
+        {
+            right = group_id;
+        }
+        else
+        {
+            left = group_id;
+        }
+        group_id = index_t((left + right) / 2);
+    }
+
+    GridwiseGemm::Run(gemm_desc_ptr[group_id].a_ptr_,
+                      gemm_desc_ptr[group_id].b_ptr_,
+                      gemm_desc_ptr[group_id].ds_ptr_,
+                      gemm_desc_ptr[group_id].e_ptr_,
+                      p_shared,
+                      a_element_op,
+                      b_element_op,
+                      cde_element_op,
+                      gemm_desc_ptr[group_id].a_grid_desc_k0_m0_m1_k1_,
+                      gemm_desc_ptr[group_id].b_grid_desc_k0_n0_n1_k1_,
+                      gemm_desc_ptr[group_id].ds_grid_desc_m0_m10_m11_n0_n10_n11_,
+                      gemm_desc_ptr[group_id].e_grid_desc_m0_m10_m11_n0_n10_n11_,
+                      gemm_desc_ptr[group_id].block_2_etile_map_,
+                      integral_constant<bool, HasMainKBlockLoop>{},
+                      integral_constant<bool, HasDoubleTailKBlockLoop>{});
+#else
+    ignore = gemm_descs_const;
+    ignore = group_count;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+#endif
+}
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t K1,
+          index_t M1PerThread,
+          index_t N1PerThread,
+          index_t KPerThread,
+          typename M1N1ThreadClusterM1Xs,
+          typename M1N1ThreadClusterN1Xs,
+          typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
+          typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+          typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
+          typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector,
+          enable_if_t<
+              is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+                  is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
+              bool> = false>
+struct DeviceGroupedGemmMultipleD_Dl : public DeviceGroupedGemm<ALayout,
+                                                                BLayout,
+                                                                DsLayout,
+                                                                ELayout,
+                                                                ADataType,
+                                                                BDataType,
+                                                                DsDataType,
+                                                                EDataType,
+                                                                AElementwiseOperation,
+                                                                BElementwiseOperation,
+                                                                CDEElementwiseOperation>
+{
+    using DeviceOp                      = DeviceGroupedGemmMultipleD_Dl;
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr auto K1Number = Number<K1>{};
+
+    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
+    {
+        assert(K % K1 == 0);
+
+        const index_t K0 = K / K1;
+
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(M, PadM)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
+    {
+        assert(K % K1 == 0);
+
+        const index_t K0 = K / K1;
+
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    template <typename ELay>
+    static auto MakeEGridDescriptor_M_N(index_t M, index_t N, index_t StrideE)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideE));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                                         const std::array<index_t, NumDTensor>& NRaws,
+                                         const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return DeviceOp::MakeEGridDescriptor_M_N<DLayout>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
+    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
+    using DsGridDesc_M_N    = decltype(MakeDsGridDescriptor_M_N({}, {}, {}));
+    using EGridDesc_M_N     = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
+
+    // GridwiseGemm
+    using GridwiseGemm =
+        GridwiseGemmDlMultipleD_km_kn_mn<BlockSize,
+                                         ADataType,
+                                         AccDataType,
+                                         DsDataType,
+                                         EDataType,
+                                         AElementwiseOperation,
+                                         BElementwiseOperation,
+                                         CDEElementwiseOperation,
+                                         InMemoryDataOperationEnum::Set,
+                                         AGridDesc_K0_M_K1,
+                                         BGridDesc_K0_N_K1,
+                                         EGridDesc_M_N,
+                                         MPerBlock,
+                                         NPerBlock,
+                                         K0PerBlock,
+                                         K1,
+                                         M1PerThread,
+                                         N1PerThread,
+                                         KPerThread,
+                                         M1N1ThreadClusterM1Xs,
+                                         M1N1ThreadClusterN1Xs,
+                                         ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterArrangeOrder,
+                                         ABlockTransferSrcAccessOrder,
+                                         ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+                                         ABlockTransferSrcVectorTensorContiguousDimOrder,
+                                         ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+                                         BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterArrangeOrder,
+                                         BBlockTransferSrcAccessOrder,
+                                         BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+                                         BBlockTransferSrcVectorTensorContiguousDimOrder,
+                                         BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+                                         CThreadTransferSrcDstAccessOrder,
+                                         CThreadTransferSrcDstVectorDim,
+                                         CThreadTransferDstScalarPerVector>;
+
+    using AGridDesc_K0_M0_M1_K1 =
+        decltype(GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
+    using BGridDesc_K0_N0_N1_K1 =
+        decltype(GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(BGridDesc_K0_N_K1{}));
+    using DsGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(DsGridDesc_M_N{}));
+    using EGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(EGridDesc_M_N{}));
+
+    struct GroupedGemmBlock2ETileMap
+    {
+        using Block2ETileMap =
+            remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBlock2CTileMap(EGridDesc_M_N{}))>;
+
+        GroupedGemmBlock2ETileMap()
+        {
+            block_2_etile_map_ = GridwiseGemm::MakeDefaultBlock2CTileMap(EGridDesc_M_N{});
+            BlockStart_        = -1;
+        }
+
+        GroupedGemmBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n, ck::index_t BlockStart)
+        {
+            block_2_etile_map_ = GridwiseGemm::MakeDefaultBlock2CTileMap(e_grid_desc_m_n);
+            BlockStart_        = BlockStart;
+        }
+
+        template <typename TopIdx>
+        __host__ __device__ constexpr auto CalculateBottomIndex(const TopIdx& idx_top) const
+        {
+            return block_2_etile_map_.CalculateBottomIndex(
+                make_multi_index(idx_top[I0] - BlockStart_));
+        }
+
+        // it's actually E-Tile
+        template <typename CTileIdx, typename CTileDim>
+        __host__ __device__ bool ValidCTileIndex(const CTileIdx& c_tile_idx,
+                                                 const CTileDim& c_tile_dim) const
+        {
+            return block_2_etile_map_.ValidCTileIndex(c_tile_idx, c_tile_dim);
+        }
+
+        __host__ bool CheckValidity(const EGridDesc_M_N& e_grid_desc_m_n) const
+        {
+            return block_2_etile_map_.CheckValidity(e_grid_desc_m_n);
+        }
+
+        Block2ETileMap block_2_etile_map_;
+        ck::index_t BlockStart_;
+    };
+
+    struct GemmKernelArg
+    {
+        // pointers
+        const ADataType* a_ptr_;
+        const BDataType* b_ptr_;
+        typename GridwiseGemm::DsGridPointer ds_ptr_;
+        EDataType* e_ptr_;
+
+        // tensor descriptors for problem definiton
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        // tensor descriptors for block/thread-wise copy
+        AGridDesc_K0_M0_M1_K1 a_grid_desc_k0_m0_m1_k1_;
+        BGridDesc_K0_N0_N1_K1 b_grid_desc_k0_n0_n1_k1_;
+        DsGridDesc_M0_M10_M11_N0_N10_N11 ds_grid_desc_m0_m10_m11_n0_n10_n11_;
+        EGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11_;
+
+        // block-to-e-tile map
+        GroupedGemmBlock2ETileMap block_2_etile_map_;
+        ck::index_t BlockStart_, BlockEnd_;
+    };
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(std::vector<const void*>& p_As,
+                 std::vector<const void*>& p_Bs,
+                 std::vector<std::array<const void*, NumDTensor>>& p_Ds,
+                 std::vector<void*>& p_Es,
+                 std::vector<GemmDesc>& gemm_descs,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op}
+        {
+            grid_size_ = 0;
+
+            group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
+
+            if(!(group_count_ == ck::type_convert<ck::index_t>(p_As.size()) &&
+                 group_count_ == ck::type_convert<ck::index_t>(p_Bs.size()) &&
+                 group_count_ == ck::type_convert<ck::index_t>(p_Es.size())))
+            {
+                throw std::runtime_error("wrong! group_count_ != p_As/b/c.size");
+            }
+
+            gemm_desc_kernel_arg_.reserve(group_count_);
+
+            skipped_group_count_ = 0;
+
+            for(std::size_t i = 0; i < gemm_descs.size(); i++)
+            {
+                const index_t M = gemm_descs[i].M_;
+                const index_t N = gemm_descs[i].N_;
+                const index_t K = gemm_descs[i].K_;
+
+                a_mtx_mraw_kraw_.emplace_back(M, K);
+                b_mtx_nraw_kraw_.emplace_back(N, K);
+
+                if(M == 0)
+                {
+                    skipped_group_count_++;
+                    continue;
+                }
+
+                const index_t StrideA = gemm_descs[i].stride_A_;
+                const index_t StrideB = gemm_descs[i].stride_B_;
+                const index_t StrideE = gemm_descs[i].stride_C_;
+
+                typename GridwiseGemm::DsGridPointer p_ds_grid{};
+                DsGridDesc_M_N ds_grid_desc_m_n;
+
+                static_for<0, NumDTensor, 1>{}([&](auto j) {
+                    using DLayout   = remove_cvref_t<tuple_element_t<j.value, DsLayout>>;
+                    using DDataType = remove_cvref_t<tuple_element_t<j.value, DsDataType>>;
+
+                    p_ds_grid(j)        = static_cast<const DDataType*>(p_Ds[i][j]);
+                    ds_grid_desc_m_n(j) = DeviceOp::MakeEGridDescriptor_M_N<DLayout>(
+                        M, N, gemm_descs[i].stride_Ds_[j]);
+                });
+
+                // tensor descriptors for problem definiton
+                const auto a_grid_desc_k0_m_k1 =
+                    DeviceOp::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
+                const auto b_grid_desc_k0_n_k1 =
+                    DeviceOp::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
+                const auto e_grid_desc_m_n =
+                    DeviceOp::MakeEGridDescriptor_M_N<ELayout>(M, N, StrideE);
+
+                if(GridwiseGemm::CheckValidity(
+                       a_grid_desc_k0_m_k1, b_grid_desc_k0_n_k1, e_grid_desc_m_n))
+                {
+
+                    const index_t grid_size_grp =
+                        GroupedGemmBlock2ETileMap(e_grid_desc_m_n, 0)
+                            .block_2_etile_map_.CalculateGridSize(e_grid_desc_m_n);
+
+                    const index_t BlockStart = grid_size_;
+                    const index_t BlockEnd   = grid_size_ + grid_size_grp;
+
+                    grid_size_ += grid_size_grp;
+
+                    // block-to-e-tile map
+                    const auto block_2_etile_map =
+                        GroupedGemmBlock2ETileMap(e_grid_desc_m_n, BlockStart);
+
+                    // tensor descriptors for block/thread-wise copy
+                    const auto a_grid_desc_k0_m0_m1_k1 =
+                        GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(a_grid_desc_k0_m_k1);
+                    const auto b_grid_desc_k0_n0_n1_k1 =
+                        GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(b_grid_desc_k0_n_k1);
+                    const auto ds_grid_desc_m0_m10_m11_n0_n10_n11 =
+                        GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(ds_grid_desc_m_n);
+                    const auto e_grid_desc_m0_m10_m11_n0_n10_n11 =
+                        GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(e_grid_desc_m_n);
+
+                    gemm_desc_kernel_arg_.push_back(
+                        GemmKernelArg{static_cast<const ADataType*>(p_As[i]),
+                                      static_cast<const BDataType*>(p_Bs[i]),
+                                      p_ds_grid,
+                                      static_cast<EDataType*>(p_Es[i]),
+                                      a_grid_desc_k0_m_k1,
+                                      b_grid_desc_k0_n_k1,
+                                      ds_grid_desc_m_n,
+                                      e_grid_desc_m_n,
+                                      a_grid_desc_k0_m0_m1_k1,
+                                      b_grid_desc_k0_n0_n1_k1,
+                                      ds_grid_desc_m0_m10_m11_n0_n10_n11,
+                                      e_grid_desc_m0_m10_m11_n0_n10_n11,
+                                      block_2_etile_map,
+                                      BlockStart,
+                                      BlockEnd});
+                }
+            }
+        }
+
+        //  private:
+        index_t group_count_;
+        index_t skipped_group_count_;
+
+        // TODO: A,B element op is unused since gridwise_gemm_dl_v1r3 does NOT support prologue
+        //       for the time being.
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+
+        std::vector<GemmKernelArg> gemm_desc_kernel_arg_;
+        std::vector<Tuple<index_t, index_t>> a_mtx_mraw_kraw_;
+        std::vector<Tuple<index_t, index_t>> b_mtx_nraw_kraw_;
+
+        index_t grid_size_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            auto K0 = arg.gemm_desc_kernel_arg_[0].a_grid_desc_k0_m_k1_.GetLength(I0);
+            bool all_has_main_k_block_loop = GridwiseGemm::CalculateHasMainKBlockLoop(K0);
+            bool all_has_double_tail_k_block_loop =
+                GridwiseGemm::CalculateHasDoubleTailKBlockLoop(K0);
+
+            for(std::size_t i = 0; i < arg.gemm_desc_kernel_arg_.size(); i++)
+            {
+#if DEBUG_LOG
+                std::cout << "group: " << i << " arg.a_grid_desc_k0_m_k1_{"
+                          << arg.gemm_desc_kernel_arg_[i].a_grid_desc_k0_m_k1_.GetLength(I0) << ", "
+                          << arg.gemm_desc_kernel_arg_[i].a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
+                          << arg.gemm_desc_kernel_arg_[i].a_grid_desc_k0_m_k1_.GetLength(I2) << "}"
+                          << std::endl;
+
+                std::cout << ", arg.b_grid_desc_k0_n_k1_{"
+                          << arg.gemm_desc_kernel_arg_[i].b_grid_desc_k0_n_k1_.GetLength(I0) << ", "
+                          << arg.gemm_desc_kernel_arg_[i].b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
+                          << arg.gemm_desc_kernel_arg_[i].b_grid_desc_k0_n_k1_.GetLength(I2) << "}"
+                          << std::endl;
+
+                std::cout << ", arg.e_grid_desc_m_n_{ "
+                          << arg.gemm_desc_kernel_arg_[i].e_grid_desc_m_n_.GetLength(I0) << ", "
+                          << arg.gemm_desc_kernel_arg_[i].e_grid_desc_m_n_.GetLength(I1) << "}"
+                          << std::endl;
+#endif
+
+                if(!GridwiseGemm::CheckValidity(arg.gemm_desc_kernel_arg_[i].a_grid_desc_k0_m_k1_,
+                                                arg.gemm_desc_kernel_arg_[i].b_grid_desc_k0_n_k1_,
+                                                arg.gemm_desc_kernel_arg_[i].e_grid_desc_m_n_))
+                {
+                    throw std::runtime_error(
+                        "wrong! GridwiseGemmDlMultipleD_km_kn_mn has invalid setting");
+                }
+
+                K0 = arg.gemm_desc_kernel_arg_[i].a_grid_desc_k0_m0_m1_k1_.GetLength(I0);
+                bool not_all_has_main_k_block_loop_same =
+                    all_has_main_k_block_loop xor GridwiseGemm::CalculateHasMainKBlockLoop(K0);
+                bool not_all_has_double_tail_k_block_loop_same =
+                    all_has_double_tail_k_block_loop xor
+                    GridwiseGemm::CalculateHasDoubleTailKBlockLoop(K0);
+
+                if(not_all_has_main_k_block_loop_same or not_all_has_double_tail_k_block_loop_same)
+                {
+                    std::ostringstream err;
+                    err << "Not all gemms have same value for [main|double_tail]_k_block_loop! in "
+                        << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
+                    throw std::runtime_error(err.str());
+                }
+            }
+
+            hipGetErrorString(
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.gemm_desc_kernel_arg_.data(),
+                                    arg.gemm_desc_kernel_arg_.size() * sizeof(GemmKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
+
+            auto launch_kernel = [&](auto has_main_k_block_loop,
+                                     auto has_double_tail_k_block_loop) {
+                constexpr bool has_main_loop   = has_main_k_block_loop.value;
+                constexpr bool has_double_loop = has_double_tail_k_block_loop.value;
+
+                const auto kernel = kernel_grouped_gemm_multiple_d_dl<GridwiseGemm,
+                                                                      GemmKernelArg,
+                                                                      AElementwiseOperation,
+                                                                      BElementwiseOperation,
+                                                                      CDEElementwiseOperation,
+                                                                      has_main_loop,
+                                                                      has_double_loop>;
+
+                return launch_and_time_kernel(
+                    stream_config,
+                    kernel,
+                    dim3(arg.grid_size_),
+                    dim3(BlockSize),
+                    0,
+                    cast_pointer_to_constant_address_space(arg.p_workspace_),
+                    arg.gemm_desc_kernel_arg_.size(),
+                    arg.a_element_op_,
+                    arg.b_element_op_,
+                    arg.cde_element_op_);
+            };
+
+            if(all_has_main_k_block_loop && all_has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, true>{});
+            }
+            else if(all_has_main_k_block_loop && !all_has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, false>{});
+            }
+            else if(!all_has_main_k_block_loop && all_has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if((ck::type_convert<ck::index_t>(arg.gemm_desc_kernel_arg_.size()) +
+            arg.skipped_group_count_) != arg.group_count_)
+        {
+            return false;
+        }
+
+        const std::string device_name = ck::get_device_name();
+
+        //  TODO add newer Navi arch
+        if(device_name != "gfx906" and device_name != "gfx908" and device_name != "gfx90a" and
+           device_name != "gfx1030")
+        {
+            return false;
+        }
+
+        for(std::size_t i = 0; i < arg.gemm_desc_kernel_arg_.size(); i++)
+        {
+            if(!GridwiseGemm::CheckValidity(arg.gemm_desc_kernel_arg_[i].a_grid_desc_k0_m_k1_,
+                                            arg.gemm_desc_kernel_arg_[i].b_grid_desc_k0_n_k1_,
+                                            arg.gemm_desc_kernel_arg_[i].e_grid_desc_m_n_))
+            {
+                return false;
+            }
+        }
+
+        return true;
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(std::vector<const void*>& p_As,
+                             std::vector<const void*>& p_Bs,
+                             std::vector<std::array<const void*, NumDTensor>>& p_Ds,
+                             std::vector<void*>& p_Es,
+                             std::vector<GemmDesc> gemm_descs,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{
+            p_As, p_Bs, p_Ds, p_Es, gemm_descs, a_element_op, b_element_op, cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::vector<const void*>& p_As,
+                        std::vector<const void*>& p_Bs,
+                        std::vector<std::array<const void*, NumDTensor>>& p_Ds,
+                        std::vector<void*>& p_Es,
+                        std::vector<GemmDesc>& gemm_descs,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(
+            p_As, p_Bs, p_Ds, p_Es, gemm_descs, a_element_op, b_element_op, cde_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedGemmMultipleD_Dl"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << M1PerThread << ", "
+            << N1PerThread << ", "
+            << KPerThread
+            << getGemmSpecializationString(GemmSpec)
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+
+    size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
+    {
+        return dynamic_cast<const Argument*>(p_arg)->group_count_ * sizeof(GemmKernelArg);
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 946a757ceeeb39040951e16f7c25540a2932a4df..ab6d2716c0445262f0cec5d53a5dafc3ab3c9555 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -43,7 +43,8 @@ __global__ void
             const B1ElementwiseOperation b1_element_op,
             const CElementwiseOperation c_element_op)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
@@ -610,10 +611,12 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
                 some_has_main_k_block_loop |= y;
             }
 
-            hipGetErrorString(hipMemcpy(arg.p_workspace_,
-                                        arg.group_kernel_args_.data(),
-                                        arg.group_kernel_args_.size() * sizeof(GroupKernelArg),
-                                        hipMemcpyHostToDevice));
+            hipGetErrorString(
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.group_kernel_args_.data(),
+                                    arg.group_kernel_args_.size() * sizeof(GroupKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
 
             float ave_time = 0;
 
@@ -678,7 +681,9 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
@@ -864,7 +869,16 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
             << "B0Spec" << getTensorSpecializationString(BSpec) << ", "
             << "B1Spec" << getTensorSpecializationString(B1Spec) << ", "
             << "CSpec" << getTensorSpecializationString(CSpec) << ", "
-            << getMaskingSpecializationString(MaskingSpec) << ">";
+            << getMaskingSpecializationString(MaskingSpec) << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle
+            << ">";
         // clang-format on
 
         return str.str();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index 07009a9e3885202a6951275b02642133e5c71889..bc9b3a4dcc15e1738c08177b2281f566617d394e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -1,6 +1,6 @@
 #pragma once
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -38,7 +38,8 @@ __global__ void
                                 const BElementwiseOperation b_element_op,
                                 const CDEElementwiseOperation c_element_op)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
@@ -381,6 +382,9 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
                 const index_t N = gemm_descs[i].N_;
                 const index_t K = gemm_descs[i].K_;
 
+                a_mtx_mraw_kraw_.emplace_back(M, K);
+                b_mtx_nraw_kraw_.emplace_back(N, K);
+
                 if(M == 0)
                 {
                     skipped_group_count_++;
@@ -485,6 +489,8 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
         CDEElementwiseOperation c_element_op_;
 
         std::vector<GemmBiasTransKernelArg> gemm_desc_kernel_arg_;
+        std::vector<Tuple<index_t, index_t>> a_mtx_mraw_kraw_;
+        std::vector<Tuple<index_t, index_t>> b_mtx_nraw_kraw_;
 
         index_t grid_size_;
     };
@@ -542,11 +548,12 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
                 }
             }
 
-            hipGetErrorString(
-                hipMemcpy(arg.p_workspace_,
-                          arg.gemm_desc_kernel_arg_.data(),
-                          arg.gemm_desc_kernel_arg_.size() * sizeof(GemmBiasTransKernelArg),
-                          hipMemcpyHostToDevice));
+            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
+                                                  arg.gemm_desc_kernel_arg_.data(),
+                                                  arg.gemm_desc_kernel_arg_.size() *
+                                                      sizeof(GemmBiasTransKernelArg),
+                                                  hipMemcpyHostToDevice,
+                                                  stream_config.stream_id_));
 
             float ave_time = 0;
 
@@ -599,7 +606,28 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
             return false;
         }
 
-        return true;
+        bool supported = true;
+
+        // If we use padding we do not support vector loads for dimensions not divisible by vector
+        // load size.
+        if constexpr(GemmSpec != GemmSpecialization::Default)
+        {
+            // [A|B]BlockTransferSrcVectorDim value define dimension in the block {K0,M,K1} layout,
+            // thus we have to adapt it to the {M,K} or {N,K} layout.
+            const auto a_raw_vector_dim = ABlockTransferSrcVectorDim != 1 ? 1 : 0;
+            const auto b_raw_vector_dim = BBlockTransferSrcVectorDim != 1 ? 1 : 0;
+
+            for(index_t i = 0; i < arg.group_count_; ++i)
+            {
+                const auto a_vector_dim = arg.a_mtx_mraw_kraw_[i].At(Number<a_raw_vector_dim>{});
+                const auto b_vector_dim = arg.b_mtx_nraw_kraw_[i].At(Number<b_raw_vector_dim>{});
+
+                supported = supported & (a_vector_dim % ABlockTransferSrcScalarPerVector == 0);
+                supported = supported & (b_vector_dim % BBlockTransferSrcScalarPerVector == 0);
+            }
+        }
+
+        return supported;
     }
 
     // polymorphic
@@ -661,7 +689,12 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
             << MPerXDL << ", "
             << NPerXDL << ", "
             << MXdlPerWave << ", "
-            << NXdlPerWave
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << getGemmSpecializationString(GemmSpec)
             << ">";
         // clang-format on
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..74f38b9db2156fe4cfb7fb43a6ce06614c030333
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -0,0 +1,623 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/ck.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/hip_check_error.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename GridwiseGemm,
+          typename GemmDesc,
+          bool HasMainKBlockLoop,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
+                                       const index_t group_count)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
+    __shared__ uint8_t p_shared[shared_size];
+
+    const index_t block_id = get_block_1d_id();
+    const auto gemm_desc_ptr =
+        reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
+
+    index_t left     = 0;
+    index_t right    = group_count;
+    index_t group_id = index_t((left + right) / 2);
+    while((!(block_id >= gemm_desc_ptr[group_id].block_start_ &&
+             block_id < gemm_desc_ptr[group_id].block_end_)) &&
+          left <= right)
+    {
+        if(block_id < gemm_desc_ptr[group_id].block_start_)
+        {
+            right = group_id;
+        }
+        else
+        {
+            left = group_id;
+        }
+        group_id = index_t((left + right) / 2);
+    }
+
+    GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
+        gemm_desc_ptr[group_id].karg_,
+        static_cast<void*>(p_shared),
+        gemm_desc_ptr[group_id].block_2_ctile_map_);
+#else
+    ignore = gemm_descs_const;
+    ignore = group_count;
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          ck::index_t NumGemmKPrefetchStage,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t KPerBlock,
+          ck::index_t AK1,
+          ck::index_t BK1,
+          ck::index_t MPerXDL,
+          ck::index_t NPerXDL,
+          ck::index_t MXdlPerWave,
+          ck::index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched = make_default_loop_scheduler(),
+          // Current implementation does not support multiple D fusions.
+          enable_if_t<AK1 == BK1 && is_same_v<DsLayout, ck::Tuple<>> &&
+                          is_same_v<DsDataType, ck::Tuple<>>,
+                      bool> = false>
+struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayout,
+                                                                           BLayout,
+                                                                           DsLayout,
+                                                                           ELayout,
+                                                                           ADataType,
+                                                                           BDataType,
+                                                                           DsDataType,
+                                                                           EDataType,
+                                                                           AElementwiseOperation,
+                                                                           BElementwiseOperation,
+                                                                           CDEElementwiseOperation>
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static_assert(KPerBlock % AK1 == 0);
+    static constexpr index_t K0PerBlock = KPerBlock / AK1;
+
+    using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
+        BlockSize,
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        EDataType,
+        ALayout,
+        BLayout,
+        ELayout,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        GemmSpec,
+        NumGemmKPrefetchStage,
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerXDL,
+        NPerXDL,
+        AK1,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsExtraN,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        CDEBlockTransferScalarPerVector_NPerBlock,
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        LoopSched,
+        PipelineVersion::v2>;
+
+    using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
+    using Block2ETileMapKSplit =
+        BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
+    // Block2CTileMap configuration parameter.
+    static constexpr index_t B2E_M01 = 8;
+    using GroupedGemmBlock2ETileMap  = OffsettedBlockToCTileMap<Block2ETileMapKSplit>;
+    using KernelArgument             = typename GridwiseGemm::Argument;
+
+    struct GemmTransKernelArg
+    {
+        KernelArgument karg_;
+        GroupedGemmBlock2ETileMap block_2_ctile_map_;
+        index_t block_start_, block_end_;
+
+        GemmTransKernelArg() = default;
+        GemmTransKernelArg(KernelArgument&& karg,
+                           GroupedGemmBlock2ETileMap&& b2c_map,
+                           index_t block_start,
+                           index_t block_end)
+            : karg_{karg},
+              block_2_ctile_map_{b2c_map},
+              block_start_{block_start},
+              block_end_{block_end}
+        {
+        }
+    };
+
+    static constexpr index_t DefaultKBatch = 1;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+
+        Argument(std::vector<const void*>& p_As,
+                 std::vector<const void*>& p_Bs,
+                 std::vector<void*>& p_Es,
+                 std::vector<GemmDesc>& gemm_descs)
+            : Argument(p_As, p_Bs, p_Es, gemm_descs, DefaultKBatch)
+        {
+            // TODO: use occupancy api to calculate appropriate batch size.
+        }
+
+        Argument(std::vector<const void*>& p_As,
+                 std::vector<const void*>& p_Bs,
+                 std::vector<void*>& p_Es,
+                 std::vector<GemmDesc>& gemm_descs,
+                 index_t kbatch)
+            : K_BATCH{kbatch}
+        {
+            grid_size_   = 0;
+            group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
+
+            if(!(group_count_ == ck::type_convert<ck::index_t>(p_As.size()) &&
+                 group_count_ == ck::type_convert<ck::index_t>(p_Bs.size()) &&
+                 group_count_ == ck::type_convert<ck::index_t>(p_Es.size())))
+            {
+                throw std::runtime_error("wrong! group_count_ != p_As/b/c.size");
+            }
+
+            gemm_kernel_args_.reserve(group_count_);
+
+            skipped_group_count_ = 0;
+
+            for(std::size_t i = 0; i < gemm_descs.size(); ++i)
+            {
+                const index_t M = gemm_descs[i].M_;
+                const index_t N = gemm_descs[i].N_;
+                const index_t K = gemm_descs[i].K_;
+
+                if(M == 0)
+                {
+                    skipped_group_count_++;
+                    continue;
+                }
+
+                const index_t stride_a = gemm_descs[i].stride_A_;
+                const index_t stride_b = gemm_descs[i].stride_B_;
+                const index_t stride_c = gemm_descs[i].stride_C_;
+
+                const index_t m_padded = GridwiseGemm::CalculateMPadded(M);
+                const index_t n_padded = GridwiseGemm::CalculateNPadded(N);
+                const index_t k_padded = GridwiseGemm::CalculateKPadded(K, K_BATCH);
+                const index_t k0       = GridwiseGemm::CalculateK0(K, K_BATCH);
+
+                const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, stride_c);
+
+                const auto local_b2c_tile_map =
+                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+                const index_t block_start = grid_size_;
+                const index_t block_end   = grid_size_ + grid_size_grp;
+
+                grid_size_ += grid_size_grp;
+
+                // block-to-e-tile map
+                auto grouped_block_2_ctile_map =
+                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
+
+                auto karg = KernelArgument{type_convert<const ADataType*>(p_As[i]),
+                                           type_convert<const BDataType*>(p_Bs[i]),
+                                           type_convert<EDataType*>(p_Es[i]),
+                                           M,
+                                           N,
+                                           K,
+                                           stride_a,
+                                           stride_b,
+                                           stride_c,
+                                           m_padded,
+                                           n_padded,
+                                           k_padded,
+                                           k0,
+                                           K_BATCH};
+
+                gemm_kernel_args_.emplace_back(
+                    std::move(karg), std::move(grouped_block_2_ctile_map), block_start, block_end);
+            }
+        }
+
+        /**
+         * @brief      Recalculate group grid size for all gemms and update B2C maps.
+         *
+         * @param[in]  kbatch  The new splitK parameter value.
+         */
+        void UpdateKBatch(index_t kbatch)
+        {
+            K_BATCH    = kbatch;
+            grid_size_ = 0;
+
+            for(std::size_t i = 0; i < gemm_kernel_args_.size(); ++i)
+            {
+
+                auto& karg = gemm_kernel_args_[i].karg_;
+
+                const index_t k_padded = GridwiseGemm::CalculateKPadded(karg.K, K_BATCH);
+                const index_t k0       = GridwiseGemm::CalculateK0(karg.K, K_BATCH);
+
+                const auto c_grid_desc_m_n =
+                    GridwiseGemm::MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
+
+                const auto local_b2c_tile_map =
+                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+                const index_t block_start = grid_size_;
+                const index_t block_end   = grid_size_ + grid_size_grp;
+
+                grid_size_ += grid_size_grp;
+
+                // block-to-e-tile map
+                auto grouped_block_2_ctile_map =
+                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
+
+                karg.KPadded                            = k_padded;
+                karg.K0                                 = k0;
+                karg.k_batch                            = K_BATCH;
+                gemm_kernel_args_[i].block_2_ctile_map_ = grouped_block_2_ctile_map;
+                gemm_kernel_args_[i].block_start_       = block_start;
+                gemm_kernel_args_[i].block_end_         = block_end;
+            }
+        }
+
+        //  private:
+        index_t K_BATCH;
+        index_t group_count_;
+        index_t skipped_group_count_;
+
+        std::vector<GemmTransKernelArg> gemm_kernel_args_;
+        index_t grid_size_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            index_t K0                       = arg.gemm_kernel_args_[0].karg_.K0;
+            bool all_have_kbatch_gt_one      = arg.gemm_kernel_args_[0].karg_.k_batch > 1;
+            bool all_have_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
+
+            for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
+            {
+                const auto& karg = arg.gemm_kernel_args_[i].karg_;
+                if(stream_config.log_level_ > 0)
+                {
+                    karg.Print();
+                }
+
+                auto kbatch = karg.k_batch;
+
+                if(!GridwiseGemm::CheckValidity(karg))
+                {
+                    std::ostringstream err;
+                    err << "Group id: " << i << " has invalid GridwiseGemm settings!" << __FILE__
+                        << ":" << __LINE__ << ", in function: " << __func__;
+                    throw std::runtime_error(err.str());
+                }
+
+                K0 = karg.K0;
+                bool not_all_have_main_k0_block_loop_same =
+                    all_have_main_k0_block_loop xor GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
+                bool not_all_have_kbatch_value_same = all_have_kbatch_gt_one xor (kbatch > 1);
+
+                if(not_all_have_main_k0_block_loop_same)
+                {
+                    std::ostringstream err;
+                    err << "Not all gemms have same value for main_k0_block_loop! in " << __FILE__
+                        << ":" << __LINE__ << ", in function: " << __func__;
+                    throw std::runtime_error(err.str());
+                }
+
+                if(not_all_have_kbatch_value_same)
+                {
+                    std::ostringstream err;
+                    err << "Not all gemms have same kbatch value (=1 or >1)! "
+                        << "group [" << i << "], kbatch: " << kbatch
+                        << ", group [0], kbatch: " << arg.gemm_kernel_args_[0].karg_.k_batch
+                        << " in " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
+                    throw std::runtime_error(err.str());
+                }
+            }
+
+            hip_check_error(
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.gemm_kernel_args_.data(),
+                                    arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
+
+            float ave_time = 0;
+
+            const auto Run = [&](const auto& kernel) {
+                if(all_have_kbatch_gt_one)
+                {
+                    for(const auto& trans_arg : arg.gemm_kernel_args_)
+                    {
+                        const auto& karg = trans_arg.karg_;
+                        hip_check_error(
+                            hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(EDataType)));
+                    }
+                }
+
+                ave_time =
+                    launch_and_time_kernel(stream_config,
+                                           kernel,
+                                           dim3(arg.grid_size_),
+                                           dim3(BlockSize),
+                                           0,
+                                           cast_pointer_to_constant_address_space(arg.p_workspace_),
+                                           arg.gemm_kernel_args_.size());
+            };
+
+            if(all_have_main_k0_block_loop)
+            {
+                if(all_have_kbatch_gt_one)
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       true,
+                                                       InMemoryDataOperationEnum::AtomicAdd>;
+
+                    Run(kernel);
+                }
+                else
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       true,
+                                                       InMemoryDataOperationEnum::Set>;
+
+                    Run(kernel);
+                }
+            }
+            else
+            {
+                if(all_have_kbatch_gt_one)
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       false,
+                                                       InMemoryDataOperationEnum::AtomicAdd>;
+
+                    Run(kernel);
+                }
+                else
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       false,
+                                                       InMemoryDataOperationEnum::Set>;
+
+                    Run(kernel);
+                }
+            }
+
+            return ave_time;
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
+            arg.skipped_group_count_) != arg.group_count_)
+        {
+#if DEBUG_LOG
+            std::cout << "The group count is not equal to sum of skipped groups "
+                         "and kernel args size!"
+                      << std::endl;
+#endif // DEBUG_LOG
+            return false;
+        }
+
+        bool supported = true;
+        for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
+        {
+            const auto& a        = arg.gemm_kernel_args_[i].karg_;
+            bool group_arg_valid = GridwiseGemm::CheckValidity(a);
+            if(not group_arg_valid)
+            {
+#if DEBUG_LOG
+                std::cout << "[" << __func__ << "] group id: " << i
+                          << " has invalid GridwiseGemm settings!" << std::endl;
+                a.Print();
+#endif // DEBUG_LOG
+            }
+            supported = supported && group_arg_valid;
+        }
+        return supported;
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(std::vector<const void*>& p_As,
+                             std::vector<const void*>& p_Bs,
+                             std::vector<std::array<const void*, NumDTensor>>&,
+                             std::vector<void*>& p_Es,
+                             std::vector<GemmDesc> gemm_descs,
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CDEElementwiseOperation)
+    {
+        return Argument{p_As, p_Bs, p_Es, gemm_descs};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::vector<const void*>& p_As,
+                        std::vector<const void*>& p_Bs,
+                        std::vector<std::array<const void*, NumDTensor>>&,
+                        std::vector<void*>& p_Es,
+                        std::vector<GemmDesc>& gemm_descs,
+                        AElementwiseOperation,
+                        BElementwiseOperation,
+                        CDEElementwiseOperation) override
+    {
+        return std::make_unique<Argument>(p_As, p_Bs, p_Es, gemm_descs);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedGemm_XdlSplitK"
+            << "<"
+            << std::string(ALayout::name)[0] << ","
+            << std::string(BLayout::name)[0] << ","
+            << std::string(ELayout::name)[0] << ","
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << getGemmSpecializationString(GemmSpec)
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+
+    size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
+    {
+        return dynamic_cast<const Argument*>(p_arg)->gemm_kernel_args_.size() *
+               sizeof(GemmTransKernelArg);
+    }
+
+    static void SetKBatchSize(Argument& arg, index_t kbatch) { arg.UpdateKBatch(kbatch); }
+
+    // polymorphic
+    void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const override
+    {
+        return SetKBatchSize(*dynamic_cast<Argument*>(p_arg), kbatch);
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..175994d49f3a3121e2e32a4ca2ef07e60c0e52e5
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp
@@ -0,0 +1,316 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// output[indices] = input
+template <typename DOutDataType,
+          typename IndexDataType,
+          typename DInDataType,
+          ck::index_t InOutVectorSize>
+struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDataType, DInDataType>
+{
+    using DInDataType_AutomicAddPreCast =
+        conditional_t<is_same_v<DInDataType, float> || is_same_v<DInDataType, double>,
+                      DInDataType,
+                      float>;
+
+    using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+    using UnaryConvert = ck::tensor_operation::element_wise::UnaryConvert;
+
+    static constexpr auto I0 = Number<0>{};
+
+    template <typename Desc_M>
+    static auto PadDescriptor_M_1d(Desc_M desc_m, index_t loop_step)
+    {
+        const auto m   = desc_m.GetLength(I0);
+        const auto pad = math::integer_least_multiple(m, loop_step) - m;
+        const auto desc_m_pad =
+            transform_tensor_descriptor(desc_m,
+                                        make_tuple(make_right_pad_transform(m, pad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+        return desc_m_pad;
+    }
+
+    static auto MakeDescriptor_M(index_t length, index_t loop_step)
+    {
+        const auto desc_m = make_naive_tensor_descriptor_packed(make_tuple(length));
+        return PadDescriptor_M_1d(desc_m, loop_step);
+    }
+
+    using InOutGrid1dDesc = decltype(MakeDescriptor_M(1, 1));
+
+    using GridwisePutElementSet = GridwisePutElement_1D<InOutGrid1dDesc,
+                                                        DOutDataType,
+                                                        IndexDataType,
+                                                        DInDataType,
+                                                        PassThrough,
+                                                        InMemoryDataOperationEnum::Set,
+                                                        InOutVectorSize>;
+
+    using GridwisePutElementAtomicAdd = GridwisePutElement_1D<InOutGrid1dDesc,
+                                                              DOutDataType,
+                                                              IndexDataType,
+                                                              DInDataType_AutomicAddPreCast,
+                                                              PassThrough,
+                                                              InMemoryDataOperationEnum::AtomicAdd,
+                                                              InOutVectorSize>;
+
+    using GridwiseCasting = GridwiseElementwise_1D<Tuple<InOutGrid1dDesc>,
+                                                   Tuple<InOutGrid1dDesc>,
+                                                   Tuple<const DInDataType_AutomicAddPreCast*>,
+                                                   Tuple<DInDataType*>,
+                                                   UnaryConvert,
+                                                   InOutVectorSize,
+                                                   Sequence<InOutVectorSize>,
+                                                   Sequence<InOutVectorSize>>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const DOutDataType* p_dout,
+                 const IndexDataType* p_indices,
+                 DInDataType* p_din,
+                 index_t dout_length,
+                 index_t din_length,
+                 const std::vector<ck::index_t>& window_lengths,
+                 const std::vector<ck::index_t>& window_strides)
+            : p_dout_{p_dout},
+              p_indices_{p_indices},
+              p_din_{p_din},
+              dout_length_raw_{dout_length},
+              din_length_raw_{din_length},
+              blockSize_{256},
+              windowOverlap_{false}
+        {
+            for(size_t i = 0; i < window_lengths.size(); ++i)
+            {
+                windowOverlap_ |= window_lengths.at(i) > window_strides.at(i);
+            }
+        }
+
+        const DOutDataType* p_dout_;
+        const IndexDataType* p_indices_;
+        DInDataType* p_din_;
+        index_t dout_length_raw_;
+        index_t din_length_raw_;
+        index_t blockSize_;
+        bool windowOverlap_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            index_t gridSize               = getAvailableComputeUnitCount(stream_config);
+            index_t loop_step              = gridSize * arg.blockSize_ * InOutVectorSize;
+            InOutGrid1dDesc din_grid_desc  = MakeDescriptor_M(arg.din_length_raw_, loop_step);
+            InOutGrid1dDesc dout_grid_desc = MakeDescriptor_M(arg.dout_length_raw_, loop_step);
+
+            if constexpr(is_same_v<DInDataType, float> || is_same_v<DInDataType, double>)
+            {
+                hip_check_error(hipMemsetAsync(arg.p_din_,
+                                               0,
+                                               arg.din_length_raw_ * sizeof(DInDataType),
+                                               stream_config.stream_id_));
+
+                if(arg.windowOverlap_)
+                {
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementAtomicAdd,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType,
+                                                                  PassThrough>;
+
+                    return launch_and_time_kernel(stream_config,
+                                                  put_kernel,
+                                                  dim3(gridSize),
+                                                  dim3(arg.blockSize_),
+                                                  0,
+                                                  dout_grid_desc,
+                                                  arg.p_dout_,
+                                                  arg.p_indices_,
+                                                  arg.p_din_,
+                                                  PassThrough{});
+                }
+                else
+                {
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementSet,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType,
+                                                                  PassThrough>;
+
+                    return launch_and_time_kernel(stream_config,
+                                                  put_kernel,
+                                                  dim3(gridSize),
+                                                  dim3(arg.blockSize_),
+                                                  0,
+                                                  dout_grid_desc,
+                                                  arg.p_dout_,
+                                                  arg.p_indices_,
+                                                  arg.p_din_,
+                                                  PassThrough{});
+                }
+            }
+            else
+            {
+                if(arg.windowOverlap_)
+                {
+                    if(arg.p_workspace_ == nullptr)
+                        throw std::runtime_error("wrong! WorkSpace pointer has not been set");
+
+                    hip_check_error(
+                        hipMemsetAsync(arg.p_workspace_,
+                                       0,
+                                       arg.din_length_raw_ * sizeof(DInDataType_AutomicAddPreCast),
+                                       stream_config.stream_id_));
+
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementAtomicAdd,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType_AutomicAddPreCast,
+                                                                  PassThrough>;
+
+                    const auto cast_kernel =
+                        kernel_elementwise_1d<GridwiseCasting,
+                                              Tuple<InOutGrid1dDesc>,
+                                              Tuple<InOutGrid1dDesc>,
+                                              Tuple<const DInDataType_AutomicAddPreCast*>,
+                                              Tuple<DInDataType*>,
+                                              UnaryConvert>;
+
+                    float elapsed_time = launch_and_time_kernel(
+                        stream_config,
+                        put_kernel,
+                        dim3(gridSize),
+                        dim3(arg.blockSize_),
+                        0,
+                        dout_grid_desc,
+                        arg.p_dout_,
+                        arg.p_indices_,
+                        static_cast<DInDataType_AutomicAddPreCast*>(arg.p_workspace_),
+                        PassThrough{});
+
+                    elapsed_time += launch_and_time_kernel(
+                        stream_config,
+                        cast_kernel,
+                        dim3(gridSize),
+                        dim3(arg.blockSize_),
+                        0,
+                        ck::make_tuple(din_grid_desc),
+                        ck::make_tuple(din_grid_desc),
+                        static_cast<DInDataType_AutomicAddPreCast*>(arg.p_workspace_),
+                        arg.p_din_,
+                        UnaryConvert{});
+
+                    return elapsed_time;
+                }
+                else
+                {
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementSet,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType,
+                                                                  PassThrough>;
+
+                    hip_check_error(hipMemsetAsync(arg.p_din_,
+                                                   0,
+                                                   arg.din_length_raw_ * sizeof(DInDataType),
+                                                   stream_config.stream_id_));
+
+                    return launch_and_time_kernel(stream_config,
+                                                  put_kernel,
+                                                  dim3(gridSize),
+                                                  dim3(arg.blockSize_),
+                                                  0,
+                                                  dout_grid_desc,
+                                                  arg.p_dout_,
+                                                  arg.p_indices_,
+                                                  arg.p_din_,
+                                                  PassThrough{});
+                }
+            }
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        bool needCast = pArg_->windowOverlap_ &&
+                        !(is_same_v<DInDataType, float> || is_same_v<DInDataType, double>);
+
+        if(!needCast)
+            return 0;
+        else
+            return pArg_->din_length_raw_ * sizeof(DInDataType_AutomicAddPreCast);
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
+        if(pArg->din_length_raw_ % InOutVectorSize != 0 ||
+           pArg->dout_length_raw_ % InOutVectorSize != 0)
+        {
+            return false;
+        }
+        return true;
+    }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_dout,
+                        const void* p_indices,
+                        void* p_din,
+                        index_t dout_length,
+                        index_t din_length,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> window_strides) override
+    {
+        // Assume p_dout, p_indices, p_din are packed memory space, dout_length and din_length are
+        // physical size of the packed tensor
+        return std::make_unique<Argument>(static_cast<const DOutDataType*>(p_dout),
+                                          static_cast<const IndexDataType*>(p_indices),
+                                          static_cast<DInDataType*>(p_din),
+                                          dout_length,
+                                          din_length,
+                                          window_lengths,
+                                          window_strides);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp
index dbeeb980a56b34f27cef97d266845ff88c10c9bc..aec5a65ccf41eafcbe6ebdf790cc75b0adbf2966 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -73,8 +73,8 @@ struct DeviceMultipleReduceMultiBlock : public DeviceMultipleReduce<Rank,
         static_for<0, NumReduction, 1>{}([&](auto I) {
             using OutDataType = remove_cvref_t<decltype(OutDataTypeTuple{}[I])>;
             flag =
-                flag && ck::reduce::InMemoryDataOperatonSupportedOnDataType<OutMemoryDataOperation,
-                                                                            OutDataType>::value;
+                flag && ck::reduce::InMemoryDataOperationSupportedOnDataType<OutMemoryDataOperation,
+                                                                             OutDataType>::value;
         });
 
         return flag;
@@ -270,8 +270,8 @@ struct DeviceMultipleReduceMultiBlock : public DeviceMultipleReduce<Rank,
                  const std::array<index_t, NumOutputDim>& outLengths,
                  const std::array<std::array<index_t, NumOutputDim>, NumReduction>& outStridesArray,
                  const std::array<int, NumReduceDim>& reduceDims,
-                 const std::array<const void*, NumReduction>& alphas,
-                 const std::array<const void*, NumReduction>& betas,
+                 const std::array<double, NumReduction>& alphas,
+                 const std::array<double, NumReduction>& betas,
                  const void* in_dev,
                  const std::array<void*, NumReduction>& out_dev_buffers,
                  const InElementwiseOperationTuple in_elementwise_op_tuple,
@@ -286,8 +286,8 @@ struct DeviceMultipleReduceMultiBlock : public DeviceMultipleReduce<Rank,
 
             for(size_t i = 0; i < NumReduction; i++)
             {
-                alpha_values_(i) = *static_cast<const AccDataType*>(alphas[i]);
-                beta_values_(i)  = *static_cast<const AccDataType*>(betas[i]);
+                alpha_values_(i) = static_cast<AccDataType>(alphas[i]);
+                beta_values_(i)  = static_cast<AccDataType>(betas[i]);
             };
 
             in_dev_ = static_cast<const InDataType*>(in_dev);
@@ -547,8 +547,8 @@ struct DeviceMultipleReduceMultiBlock : public DeviceMultipleReduce<Rank,
         const std::array<index_t, NumOutputDim> outLengths,
         const std::array<std::array<index_t, NumOutputDim>, NumReduction> outStridesArray,
         const std::array<int, NumReduceDim> reduceDims,
-        const std::array<const void*, NumReduction> alphas,
-        const std::array<const void*, NumReduction> betas,
+        const std::array<double, NumReduction> alphas,
+        const std::array<double, NumReduction> betas,
         const void* in_dev,
         const std::array<void*, NumReduction> out_dev_buffers,
         const InElementwiseOperationTuple in_elementwise_op_tuple,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp
index ff8465e9fc736679a05b7059d0aa4e4342404757..6d1d5c8e239de8f57df1baa82d4981e8f30e54b6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -195,8 +195,8 @@ struct DeviceMultipleReduceThreadWise : public DeviceMultipleReduce<Rank,
                  const std::array<index_t, NumOutputDim>& outLengths,
                  const std::array<std::array<index_t, NumOutputDim>, NumReduction>& outStridesArray,
                  const std::array<int, NumReduceDim>& reduceDims,
-                 const std::array<const void*, NumReduction>& alphas,
-                 const std::array<const void*, NumReduction>& betas,
+                 const std::array<double, NumReduction>& alphas,
+                 const std::array<double, NumReduction>& betas,
                  const void* in_dev,
                  const std::array<void*, NumReduction>& out_dev_buffers,
                  const InElementwiseOperationTuple in_elementwise_op_tuple,
@@ -211,8 +211,8 @@ struct DeviceMultipleReduceThreadWise : public DeviceMultipleReduce<Rank,
 
             for(size_t i = 0; i < NumReduction; i++)
             {
-                alpha_values_(i) = *static_cast<const AccDataType*>(alphas[i]);
-                beta_values_(i)  = *static_cast<const AccDataType*>(betas[i]);
+                alpha_values_(i) = static_cast<AccDataType>(alphas[i]);
+                beta_values_(i)  = static_cast<AccDataType>(betas[i]);
             };
 
             in_dev_ = static_cast<const InDataType*>(in_dev);
@@ -374,8 +374,8 @@ struct DeviceMultipleReduceThreadWise : public DeviceMultipleReduce<Rank,
         const std::array<index_t, NumOutputDim> outLengths,
         const std::array<std::array<index_t, NumOutputDim>, NumReduction> outStridesArray,
         const std::array<int, NumReduceDim> reduceDims,
-        const std::array<const void*, NumReduction> alphas,
-        const std::array<const void*, NumReduction> betas,
+        const std::array<double, NumReduction> alphas,
+        const std::array<double, NumReduction> betas,
         const void* in_dev,
         const std::array<void*, NumReduction> out_dev_buffers,
         const InElementwiseOperationTuple in_elementwise_op_tuple,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
index 47d9df80255805729b6c36fed9ae11608052c51d..ea0d805043e07fe3d818fbb7bf7ef42e952496a9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -10,57 +10,25 @@
 #include "ck/tensor_operation/gpu/device/device_normalization.hpp"
 #include "ck/tensor_operation/gpu/device/device_reduce.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 
-namespace ck {
-template <typename GridwiseReduction,
-          typename XDataType,
-          typename GammaDataType,
-          typename BetaDataType,
-          typename YDataType,
-          typename AccDataType,
-          typename AccElementwiseOperation,
-          typename GridDesc_M_K>
-__global__ void kernel_normalization(const GridDesc_M_K x_grid_desc_m_k,
-                                     const GridDesc_M_K gamma_grid_desc_m_k,
-                                     const GridDesc_M_K beta_grid_desc_m_k,
-                                     const GridDesc_M_K y_grid_desc_m_k,
-                                     index_t num_k_block_tile_iteration,
-                                     AccDataType epsilon,
-                                     const XDataType* const __restrict__ p_x_global,
-                                     const GammaDataType* const __restrict__ p_gamma_global,
-                                     const BetaDataType* const __restrict__ p_beta_global,
-                                     YDataType* const __restrict__ p_y_global,
-                                     const AccElementwiseOperation acc_elementwise_op)
-{
-    GridwiseReduction::Run(x_grid_desc_m_k,
-                           gamma_grid_desc_m_k,
-                           beta_grid_desc_m_k,
-                           y_grid_desc_m_k,
-                           num_k_block_tile_iteration,
-                           epsilon,
-                           p_x_global,
-                           p_gamma_global,
-                           p_beta_global,
-                           p_y_global,
-                           acc_elementwise_op);
-};
-} // namespace ck
-
 namespace ck {
 namespace tensor_operation {
 namespace device {
 
 // Y = Normalization(X, Beta, Gamma)
+// M: Invarient length
+// K: Reduce length (Calculate mean and variance along K dimension)
+// eg. Length = [N, C, H, W], reduce dim = [C, H, W]
+// Then, M = N, K = C * H * W
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
-          typename AccDataType,
+          typename ComputeDataType,
           typename YDataType,
-          typename AccElementwiseOperation,
+          typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim,
           index_t BlockSize,
@@ -74,16 +42,18 @@ template <typename XDataType,
           index_t GammaSrcVectorSize,
           index_t BetaSrcVectorDim,
           index_t BetaSrcVectorSize,
-          index_t YDstVectorSize>
+          index_t YDstVectorSize,
+          bool UseWelford = true>
 struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                                                             GammaDataType,
                                                             BetaDataType,
-                                                            AccDataType,
+                                                            ComputeDataType,
                                                             YDataType,
-                                                            AccElementwiseOperation,
+                                                            YElementwiseOperation,
                                                             Rank,
                                                             NumReduceDim>
 {
+    static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize);
     static_assert(
         ((GammaSrcVectorDim == 0 && MThreadSliceSize % GammaSrcVectorSize == 0) ||
          (GammaSrcVectorDim == 1 && KThreadSliceSize % GammaSrcVectorSize == 0)),
@@ -101,7 +71,6 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
     static auto MakeSrc2dDescriptor(const std::vector<index_t>& inLengths,
                                     const std::vector<index_t>& inStrides,
-                                    int blkGroupSize,
                                     int numBlockTileIteration)
     {
         constexpr index_t NumInvariantDim  = Rank - NumReduceDim;
@@ -150,10 +119,9 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         const auto invariantLength = in_grid_desc_m_k.GetLength(Number<0>{});
         const auto reduceLength    = in_grid_desc_m_k.GetLength(Number<1>{});
 
-        const int reduceSizePerBlock = K_BlockTileSize * numBlockTileIteration;
         const auto inPad_M =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
-        const auto inPad_K = reduceSizePerBlock * blkGroupSize - reduceLength;
+        const auto inPad_K = K_BlockTileSize * numBlockTileIteration - reduceLength;
 
         auto in_grid_desc_m_k_padded = transform_tensor_descriptor(
             in_grid_desc_m_k,
@@ -165,52 +133,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         return (in_grid_desc_m_k_padded);
     };
 
-    using GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1));
-
-    using GridwiseReduceLayernormGeneric =
-        GridwiseNormalizationWelfordVariance_mk_to_mk<XDataType,
-                                                      GammaDataType,
-                                                      BetaDataType,
-                                                      YDataType,
-                                                      AccDataType,
-                                                      AccElementwiseOperation,
-                                                      GridDesc_M_K,
-                                                      BlockSize,
-                                                      MThreadClusterSize,
-                                                      KThreadClusterSize,
-                                                      MThreadSliceSize,
-                                                      KThreadSliceSize,
-                                                      XYSrcVectorDim,
-                                                      XSrcVectorSize,
-                                                      GammaSrcVectorDim,
-                                                      GammaSrcVectorSize,
-                                                      BetaSrcVectorDim,
-                                                      BetaSrcVectorSize,
-                                                      XYSrcVectorDim,
-                                                      YDstVectorSize,
-                                                      false>;
-    using GridwiseNormalizationSweepOnce =
-        GridwiseNormalizationWelfordVariance_mk_to_mk<XDataType,
-                                                      GammaDataType,
-                                                      BetaDataType,
-                                                      YDataType,
-                                                      AccDataType,
-                                                      AccElementwiseOperation,
-                                                      GridDesc_M_K,
-                                                      BlockSize,
-                                                      MThreadClusterSize,
-                                                      KThreadClusterSize,
-                                                      MThreadSliceSize,
-                                                      KThreadSliceSize,
-                                                      XYSrcVectorDim,
-                                                      XSrcVectorSize,
-                                                      GammaSrcVectorDim,
-                                                      GammaSrcVectorSize,
-                                                      BetaSrcVectorDim,
-                                                      BetaSrcVectorSize,
-                                                      XYSrcVectorDim,
-                                                      YDstVectorSize,
-                                                      true>;
+    using GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1));
 
     struct Argument : public BaseArgument
     {
@@ -220,51 +143,48 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                  const std::vector<index_t> betaStrides,
                  const std::vector<index_t> yStrides,
                  const std::vector<index_t> reduceDims,
-                 AccElementwiseOperation acc_elementwise_op,
-                 AccDataType epsilon,
+                 YElementwiseOperation y_elementwise_op,
+                 double epsilon,
                  const XDataType* p_x,
                  const GammaDataType* p_gamma,
                  const BetaDataType* p_beta,
                  YDataType* p_y)
-            : epsilon_(epsilon),
-              p_x_(p_x),
+            : p_x_(p_x),
               p_gamma_(p_gamma),
               p_beta_(p_beta),
               p_y_(p_y),
-              acc_elementwise_op_(acc_elementwise_op)
+              y_elementwise_op_(y_elementwise_op)
         {
+            epsilon_ = static_cast<ComputeDataType>(epsilon);
+
             Lengths_      = shuffle_tensor_dimensions<Rank, NumReduceDim>(lengths, reduceDims);
             xStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(xStrides, reduceDims);
             yStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(yStrides, reduceDims);
             gammaStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(gammaStrides, reduceDims);
             betaStrides_  = shuffle_tensor_dimensions<Rank, NumReduceDim>(betaStrides, reduceDims);
 
-            long_index_t invariant_total_length;
-            long_index_t reduce_total_length;
+            long_index_t invariant_length;
+            long_index_t reduce_length;
 
-            std::tie(invariant_total_length, reduce_total_length) =
+            std::tie(invariant_length, reduce_length) =
                 get_2d_lengths<Rank, NumReduceDim>(Lengths_);
 
-            blkGroupSize_          = 1;
-            numBlockTileIteration_ = (reduce_total_length + K_BlockTileSize - 1) / K_BlockTileSize;
+            numBlockTileIteration_ = math::integer_divide_ceil(reduce_length, K_BlockTileSize);
 
-            gridSize_ = math::integer_least_multiple(invariant_total_length, M_BlockTileSize) /
-                        M_BlockTileSize * blkGroupSize_;
+            gridSize_ = math::integer_divide_ceil(invariant_length, M_BlockTileSize);
 
-            x_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, xStrides_, blkGroupSize_, numBlockTileIteration_);
+            x_grid_desc_m_k_ = MakeSrc2dDescriptor(Lengths_, xStrides_, numBlockTileIteration_);
             gamma_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, gammaStrides_, blkGroupSize_, numBlockTileIteration_);
+                MakeSrc2dDescriptor(Lengths_, gammaStrides_, numBlockTileIteration_);
             beta_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, betaStrides_, blkGroupSize_, numBlockTileIteration_);
-            y_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, yStrides_, blkGroupSize_, numBlockTileIteration_);
+                MakeSrc2dDescriptor(Lengths_, betaStrides_, numBlockTileIteration_);
+            y_grid_desc_m_k_ = MakeSrc2dDescriptor(Lengths_, yStrides_, numBlockTileIteration_);
 
             isSweeponce_ =
                 x_grid_desc_m_k_.GetLength(Number<1>{}) <= KThreadClusterSize * KThreadSliceSize;
         }
 
-        AccDataType epsilon_;
+        ComputeDataType epsilon_;
 
         const XDataType* p_x_;
         const GammaDataType* p_gamma_;
@@ -277,9 +197,8 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         std::vector<index_t> betaStrides_;
         std::vector<index_t> yStrides_;
 
-        AccElementwiseOperation acc_elementwise_op_;
+        YElementwiseOperation y_elementwise_op_;
 
-        int blkGroupSize_;
         int numBlockTileIteration_;
         size_t gridSize_;
 
@@ -294,23 +213,27 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
     {
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
-            const auto kernel_main = arg.isSweeponce_
-                                         ? kernel_normalization<GridwiseNormalizationSweepOnce,
-                                                                XDataType,
-                                                                GammaDataType,
-                                                                BetaDataType,
-                                                                YDataType,
-                                                                AccDataType,
-                                                                AccElementwiseOperation,
-                                                                GridDesc_M_K>
-                                         : kernel_normalization<GridwiseReduceLayernormGeneric,
-                                                                XDataType,
-                                                                GammaDataType,
-                                                                BetaDataType,
-                                                                YDataType,
-                                                                AccDataType,
-                                                                AccElementwiseOperation,
-                                                                GridDesc_M_K>;
+            auto kernel_main = NormalizationKernelSelector<XDataType,
+                                                           GammaDataType,
+                                                           BetaDataType,
+                                                           YDataType,
+                                                           ComputeDataType,
+                                                           YElementwiseOperation,
+                                                           GridDesc_M_K,
+                                                           BlockSize,
+                                                           MThreadClusterSize,
+                                                           KThreadClusterSize,
+                                                           MThreadSliceSize,
+                                                           KThreadSliceSize,
+                                                           XYSrcVectorDim,
+                                                           XSrcVectorSize,
+                                                           GammaSrcVectorDim,
+                                                           GammaSrcVectorSize,
+                                                           BetaSrcVectorDim,
+                                                           BetaSrcVectorSize,
+                                                           XYSrcVectorDim,
+                                                           YDstVectorSize,
+                                                           UseWelford>(arg.isSweeponce_);
 
             float avg_time = 0;
             avg_time += launch_and_time_kernel(stream_config,
@@ -328,7 +251,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                                                arg.p_gamma_,
                                                arg.p_beta_,
                                                arg.p_y_,
-                                               arg.acc_elementwise_op_);
+                                               arg.y_elementwise_op_);
 
             return (avg_time);
         };
@@ -359,6 +282,9 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
                 if(p_arg_->invariant_lowest_length % XSrcVectorSize != 0)
                     return false;
+
+                if(p_arg_->invariant_lowest_length % YDstVectorSize != 0)
+                    return false;
             };
         }
         else
@@ -368,12 +294,12 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
             if(p_arg_->Lengths_[Rank - 1] % XSrcVectorSize != 0)
                 return false;
-        };
 
-        if(p_arg_->Lengths_[Rank - 1] % YDstVectorSize != 0)
-        {
-            return false;
-        }
+            if(p_arg_->Lengths_[Rank - 1] % YDstVectorSize != 0)
+            {
+                return false;
+            }
+        };
 
         // if fastest dim is not reduced
         if constexpr(GammaSrcVectorDim == 0)
@@ -421,14 +347,14 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                         const std::vector<index_t> betaStrides,
                         const std::vector<index_t> yStrides,
                         const std::vector<index_t> reduceDims,
-                        AccDataType epsilon,
+                        double epsilon,
                         const void* p_x,
                         const void* p_gamma,
                         const void* p_beta,
                         void* p_y,
                         void* p_saveMean,
                         void* p_saveInvVar,
-                        AccElementwiseOperation acc_elementwise_op) override
+                        YElementwiseOperation y_elementwise_op) override
     {
         // TODO
         // Optional cache of the intermediate results (mean and InvVariance) during the
@@ -442,7 +368,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                                           betaStrides,
                                           yStrides,
                                           reduceDims,
-                                          acc_elementwise_op,
+                                          y_elementwise_op,
                                           epsilon,
                                           static_cast<const XDataType*>(p_x),
                                           static_cast<const GammaDataType*>(p_gamma),
@@ -461,8 +387,8 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
         // clang-format off
         str << "DeviceNormalizationImpl<" << BlockSize << ",";
-        str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
-        str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
+        str << "Cluster_MK_" << MThreadClusterSize << "_" << KThreadClusterSize << ",";
+        str << "Slice_MK_" << MThreadSliceSize << "_" << KThreadSliceSize << ",";
         str << "XYSrcVectorDim_" << XYSrcVectorDim  << ",";
         str << "VectorSize_X" << XSrcVectorSize << "_Gamma" << GammaSrcVectorSize << "_Beta" << BetaSrcVectorSize << "_Y" << YDstVectorSize << ">";
         // clang-format on
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..8b2b3c41bfdf3f4793263dd65fc5f6dc941eb13b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
@@ -0,0 +1,658 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/reduction_operator.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/device/device_normalization.hpp"
+#include "ck/tensor_operation/gpu/device/device_reduce.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+template <typename GridwiseWelford,
+          typename XDataType,
+          typename MeanVarDataType,
+          typename ComputeDataType,
+          typename XGridDesc_M_K,
+          typename MeanVarGridDesc_M_KBlock>
+__global__ void
+kernel_normalizationSplitK1st(const XGridDesc_M_K x_grid_desc_m_k,
+                              const MeanVarGridDesc_M_KBlock mean_var_grid_desc_m_kblock,
+                              index_t num_k_block_tile_iteration,
+                              const XDataType* const __restrict__ p_x_global,
+                              MeanVarDataType* const __restrict__ p_welford_mean,
+                              MeanVarDataType* const __restrict__ p_welford_variance,
+                              int32_t* const __restrict__ p_welford_count)
+{
+    GridwiseWelford::Run(x_grid_desc_m_k,
+                         mean_var_grid_desc_m_kblock,
+                         num_k_block_tile_iteration,
+                         p_x_global,
+                         p_welford_mean,
+                         p_welford_variance,
+                         p_welford_count);
+};
+
+template <typename GridwiseWelfordNormalization,
+          typename MeanVarDataType,
+          typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
+          typename MeanVarGridDesc_M_KBlock,
+          typename CountGridDesc_M_KBlock,
+          typename XYGammaBetaGridDesc_M_K>
+__global__ void
+kernel_normalizationSplitK2nd(const MeanVarGridDesc_M_KBlock mean_var_grid_desc_m_kblock,
+                              const CountGridDesc_M_KBlock count_grid_desc_m_kblock,
+                              const XYGammaBetaGridDesc_M_K x_grid_desc_m_k,
+                              const XYGammaBetaGridDesc_M_K gamma_grid_desc_m_k,
+                              const XYGammaBetaGridDesc_M_K beta_grid_desc_m_k,
+                              const XYGammaBetaGridDesc_M_K y_grid_desc_m_k,
+                              index_t num_k_mean_var_count_iteration,
+                              index_t num_k_block_tile_iteration,
+                              index_t k_grid_size,
+                              ComputeDataType epsilon,
+                              const MeanVarDataType* const p_mean_global,
+                              const MeanVarDataType* const p_variance_global,
+                              const int32_t* const p_welford_count_global,
+                              const XDataType* const __restrict__ p_x_global,
+                              const GammaDataType* const __restrict__ p_gamma_global,
+                              const BetaDataType* const __restrict__ p_beta_global,
+                              YDataType* const __restrict__ p_y_global,
+                              const YElementwiseOperation y_elementwise_op)
+{
+    GridwiseWelfordNormalization::Run(mean_var_grid_desc_m_kblock,
+                                      count_grid_desc_m_kblock,
+                                      x_grid_desc_m_k,
+                                      gamma_grid_desc_m_k,
+                                      beta_grid_desc_m_k,
+                                      y_grid_desc_m_k,
+                                      num_k_mean_var_count_iteration,
+                                      num_k_block_tile_iteration,
+                                      k_grid_size,
+                                      epsilon,
+                                      p_mean_global,
+                                      p_variance_global,
+                                      p_welford_count_global,
+                                      p_x_global,
+                                      p_gamma_global,
+                                      p_beta_global,
+                                      p_y_global,
+                                      y_elementwise_op);
+};
+} // namespace ck
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// Y = Normalization(X, Beta, Gamma)
+// M: Invarient length
+// K: Reduce length (Calculate mean and variance along K dimension)
+// eg. Length = [N, C, H, W], reduce dim = [C, H, W]
+// Then, M = N, K = C * H * W
+template <typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename ComputeDataType,
+          typename YDataType,
+          typename YElementwiseOperation,
+          index_t Rank,
+          index_t NumReduceDim,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XYVectorDim,
+          index_t XSrcVectorSize,
+          index_t GammaSrcVectorDim,
+          index_t GammaSrcVectorSize,
+          index_t BetaSrcVectorDim,
+          index_t BetaSrcVectorSize,
+          index_t YDstVectorSize>
+struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
+                                                                  GammaDataType,
+                                                                  BetaDataType,
+                                                                  ComputeDataType,
+                                                                  YDataType,
+                                                                  YElementwiseOperation,
+                                                                  Rank,
+                                                                  NumReduceDim>
+{
+    using MeanVarDataType = ComputeDataType;
+
+    static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize);
+    static_assert(
+        ((GammaSrcVectorDim == 0 && MThreadSliceSize % GammaSrcVectorSize == 0) ||
+         (GammaSrcVectorDim == 1 && KThreadSliceSize % GammaSrcVectorSize == 0)),
+        "Invalid thread slice sizes and/or gamma vector sizes configuration, please check!");
+
+    static_assert(
+        ((BetaSrcVectorDim == 0 && MThreadSliceSize % BetaSrcVectorSize == 0) ||
+         (BetaSrcVectorDim == 1 && KThreadSliceSize % BetaSrcVectorSize == 0)),
+        "Invalid thread slice sizes and/or beta vector sizes configuration, please check!");
+
+    using PassThrough = tensor_operation::element_wise::PassThrough;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static auto MakeSrc2dDescriptor(const std::vector<index_t>& inLengths,
+                                    const std::vector<index_t>& inStrides,
+                                    int kBlockSize,
+                                    int numBlockTileIteration)
+    {
+        constexpr index_t NumInvariantDim  = Rank - NumReduceDim;
+        static constexpr index_t numSrcDim = Rank;
+        static constexpr bool reduceAllDim = (NumInvariantDim == 0);
+
+        const auto tupleSrcLengths = make_tuple_from_array(inLengths, Number<numSrcDim>{});
+        const auto tupleSrcStrides = make_tuple_from_array(inStrides, Number<numSrcDim>{});
+
+        const auto inDesc = make_naive_tensor_descriptor(tupleSrcLengths, tupleSrcStrides);
+
+        const auto in_grid_desc_m_k = [&]() {
+            if constexpr(reduceAllDim)
+            {
+                const auto one_dim_inDesc = transform_tensor_descriptor(
+                    inDesc,
+                    make_tuple(make_merge_transform(tupleSrcLengths)),
+                    make_tuple(typename arithmetic_sequence_gen<0, numSrcDim, 1>::type{}),
+                    make_tuple(Sequence<0>{}));
+
+                return transform_tensor_descriptor(one_dim_inDesc,
+                                                   make_tuple(make_unmerge_transform(make_tuple(
+                                                       1, one_dim_inDesc.GetLength(Number<0>{})))),
+                                                   make_tuple(Sequence<0>{}),
+                                                   make_tuple(Sequence<0, 1>{}));
+            }
+            else
+            {
+                using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
+                using ReduceDims = typename arithmetic_sequence_gen<NumInvariantDim, Rank, 1>::type;
+
+                const auto reduceDimLengths =
+                    make_tuple_from_array_and_index_seq(inLengths, ReduceDims{});
+                const auto invariantDimLengths =
+                    make_tuple_from_array_and_index_seq(inLengths, InvariantDims{});
+
+                return transform_tensor_descriptor(
+                    inDesc,
+                    make_tuple(make_merge_transform(invariantDimLengths),
+                               make_merge_transform(reduceDimLengths)),
+                    make_tuple(InvariantDims{}, ReduceDims{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
+            }
+        }();
+
+        const auto invariantLength = in_grid_desc_m_k.GetLength(Number<0>{});
+        const auto reduceLength    = in_grid_desc_m_k.GetLength(Number<1>{});
+
+        const int reduceSizePerBlock = K_BlockTileSize * numBlockTileIteration;
+        const auto inPad_M =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+        const auto inPad_K = reduceSizePerBlock * kBlockSize - reduceLength;
+
+        auto in_grid_desc_m_k_padded = transform_tensor_descriptor(
+            in_grid_desc_m_k,
+            make_tuple(make_right_pad_transform(invariantLength, inPad_M),
+                       make_right_pad_transform(reduceLength, inPad_K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (in_grid_desc_m_k_padded);
+    };
+
+    template <typename DoPads, index_t MPerTile, index_t KPerTile>
+    static auto MakeMeanVarDescriptor_M_K(index_t M, index_t K)
+    {
+        const auto grid_desc_m_k =
+            make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(K, I1));
+        return PadTensorDescriptor(grid_desc_m_k, make_tuple(MPerTile, KPerTile), DoPads{});
+    }
+
+    template <typename DoPads, index_t MPerTile, index_t KPerTile>
+    static auto MakeCountDescriptor_M_K(index_t M, index_t K)
+    {
+        const auto grid_desc_m_k =
+            make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I0, I1));
+        return PadTensorDescriptor(grid_desc_m_k, make_tuple(MPerTile, KPerTile), DoPads{});
+    }
+
+    using SrcGridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1));
+    using Kernel1MeanVarGridDesc_M_KBlock =
+        decltype(MakeMeanVarDescriptor_M_K<Sequence<true, false>, 1, 1>(1, 1));
+
+    using Kernel2MeanVarGridDesc_M_KBlock =
+        decltype(MakeMeanVarDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
+
+    using Kernel2CountGridDesc_M_KBlock =
+        decltype(MakeCountDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
+
+    using GridwiseWelford = GridwiseNormalizationSplitK1st<XDataType,
+                                                           ComputeDataType,
+                                                           MeanVarDataType,
+                                                           SrcGridDesc_M_K,
+                                                           Kernel1MeanVarGridDesc_M_KBlock,
+                                                           BlockSize,
+                                                           MThreadClusterSize,
+                                                           KThreadClusterSize,
+                                                           MThreadSliceSize,
+                                                           KThreadSliceSize,
+                                                           XYVectorDim,
+                                                           XSrcVectorSize>;
+
+    using GridwiseWelfordNormalization =
+        GridwiseNormalizationSplitK2nd<MeanVarDataType,
+                                       XDataType,
+                                       GammaDataType,
+                                       BetaDataType,
+                                       YDataType,
+                                       ComputeDataType,
+                                       YElementwiseOperation,
+                                       Kernel2MeanVarGridDesc_M_KBlock,
+                                       Kernel2CountGridDesc_M_KBlock,
+                                       SrcGridDesc_M_K,
+                                       BlockSize,
+                                       MThreadClusterSize,
+                                       KThreadClusterSize,
+                                       MThreadSliceSize,
+                                       KThreadSliceSize,
+                                       XYVectorDim,
+                                       XSrcVectorSize,
+                                       GammaSrcVectorDim,
+                                       GammaSrcVectorSize,
+                                       BetaSrcVectorDim,
+                                       BetaSrcVectorSize,
+                                       XYVectorDim,
+                                       YDstVectorSize>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const std::vector<index_t> lengths,
+                 const std::vector<index_t> xStrides,
+                 const std::vector<index_t> gammaStrides,
+                 const std::vector<index_t> betaStrides,
+                 const std::vector<index_t> yStrides,
+                 const std::vector<index_t> reduceDims,
+                 YElementwiseOperation y_elementwise_op,
+                 double epsilon,
+                 const XDataType* p_x,
+                 const GammaDataType* p_gamma,
+                 const BetaDataType* p_beta,
+                 YDataType* p_y)
+            : p_x_(p_x),
+              p_gamma_(p_gamma),
+              p_beta_(p_beta),
+              p_y_(p_y),
+              p_workspace_mean_{nullptr},
+              p_workspace_var_{nullptr},
+              p_workspace_count_{nullptr},
+              y_elementwise_op_(y_elementwise_op)
+        {
+            epsilon_ = static_cast<ComputeDataType>(epsilon);
+
+            Lengths_      = shuffle_tensor_dimensions<Rank, NumReduceDim>(lengths, reduceDims);
+            xStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(xStrides, reduceDims);
+            yStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(yStrides, reduceDims);
+            gammaStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(gammaStrides, reduceDims);
+            betaStrides_  = shuffle_tensor_dimensions<Rank, NumReduceDim>(betaStrides, reduceDims);
+
+            std::tie(MRaw_, KRaw_) = get_2d_lengths<Rank, NumReduceDim>(Lengths_);
+
+            numBlockTileIteration_ = 1;
+            while(true)
+            {
+                int testKGridSize =
+                    math::integer_divide_ceil(KRaw_, K_BlockTileSize * numBlockTileIteration_);
+
+                // we want the kGridSize_ be not more than 128
+                if(testKGridSize <= 128)
+                    break;
+
+                ++numBlockTileIteration_;
+            };
+
+            kGridSize_ = math::integer_divide_ceil(KRaw_, K_BlockTileSize * numBlockTileIteration_);
+            gridSize_  = math::integer_divide_ceil(MRaw_, M_BlockTileSize) * kGridSize_;
+
+            // We do not use vector load for mean, var and count
+            static constexpr index_t K_MeanVarCountBlockTileSize = KThreadClusterSize;
+
+            numMeanVarCountIteration_ =
+                math::integer_divide_ceil(kGridSize_, K_MeanVarCountBlockTileSize);
+
+            x_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, xStrides_, kGridSize_, numBlockTileIteration_);
+            gamma_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, gammaStrides_, kGridSize_, numBlockTileIteration_);
+            beta_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, betaStrides_, kGridSize_, numBlockTileIteration_);
+            y_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, yStrides_, kGridSize_, numBlockTileIteration_);
+
+            // We don't need to pad in K dimension for Welford1. Set KPerTile 1.
+            kernel1_mean_var_grid_desc_m_kblock_ =
+                MakeMeanVarDescriptor_M_K<Sequence<true, false>, M_BlockTileSize, 1>(MRaw_,
+                                                                                     kGridSize_);
+
+            kernel2_mean_var_grid_desc_m_kblock_ =
+                MakeMeanVarDescriptor_M_K<Sequence<true, true>,
+                                          M_BlockTileSize,
+                                          K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
+
+            kernel2_count_grid_desc_m_kblock_ =
+                MakeCountDescriptor_M_K<Sequence<true, true>,
+                                        M_BlockTileSize,
+                                        K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
+        }
+
+        ComputeDataType epsilon_;
+
+        const XDataType* p_x_;
+        const GammaDataType* p_gamma_;
+        const BetaDataType* p_beta_;
+        YDataType* p_y_;
+        void* p_workspace_mean_;
+        void* p_workspace_var_;
+        void* p_workspace_count_;
+
+        std::vector<index_t> Lengths_;
+        std::vector<index_t> xStrides_;
+        std::vector<index_t> gammaStrides_;
+        std::vector<index_t> betaStrides_;
+        std::vector<index_t> yStrides_;
+
+        YElementwiseOperation y_elementwise_op_;
+
+        int kGridSize_;
+        int numMeanVarCountIteration_;
+        int numBlockTileIteration_;
+        size_t gridSize_;
+
+        SrcGridDesc_M_K x_grid_desc_m_k_;
+        SrcGridDesc_M_K gamma_grid_desc_m_k_;
+        SrcGridDesc_M_K beta_grid_desc_m_k_;
+        SrcGridDesc_M_K y_grid_desc_m_k_;
+
+        Kernel1MeanVarGridDesc_M_KBlock kernel1_mean_var_grid_desc_m_kblock_;
+        Kernel2MeanVarGridDesc_M_KBlock kernel2_mean_var_grid_desc_m_kblock_;
+        Kernel2CountGridDesc_M_KBlock kernel2_count_grid_desc_m_kblock_;
+
+        index_t MRaw_; // invarient length
+        index_t KRaw_; // reduce length
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(arg.p_workspace_mean_ == nullptr || arg.p_workspace_var_ == nullptr ||
+               arg.p_workspace_count_ == nullptr)
+                throw std::runtime_error("wrong! WorkSpace pointer has not been set");
+
+            auto kernel1 = kernel_normalizationSplitK1st<GridwiseWelford,
+                                                         XDataType,
+                                                         MeanVarDataType,
+                                                         ComputeDataType,
+                                                         SrcGridDesc_M_K,
+                                                         Kernel1MeanVarGridDesc_M_KBlock>;
+
+            auto kernel2 = kernel_normalizationSplitK2nd<GridwiseWelfordNormalization,
+                                                         MeanVarDataType,
+                                                         XDataType,
+                                                         GammaDataType,
+                                                         BetaDataType,
+                                                         YDataType,
+                                                         ComputeDataType,
+                                                         YElementwiseOperation,
+                                                         Kernel2MeanVarGridDesc_M_KBlock,
+                                                         Kernel2CountGridDesc_M_KBlock,
+                                                         SrcGridDesc_M_K>;
+
+            float avg_time = 0;
+            avg_time += launch_and_time_kernel(stream_config,
+                                               kernel1,
+                                               dim3(arg.gridSize_),
+                                               dim3(BlockSize),
+                                               0,
+                                               arg.x_grid_desc_m_k_,
+                                               arg.kernel1_mean_var_grid_desc_m_kblock_,
+                                               arg.numBlockTileIteration_,
+                                               arg.p_x_,
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_mean_),
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_var_),
+                                               static_cast<int32_t*>(arg.p_workspace_count_));
+
+            avg_time += launch_and_time_kernel(stream_config,
+                                               kernel2,
+                                               dim3(arg.gridSize_),
+                                               dim3(BlockSize),
+                                               0,
+                                               arg.kernel2_mean_var_grid_desc_m_kblock_,
+                                               arg.kernel2_count_grid_desc_m_kblock_,
+                                               arg.x_grid_desc_m_k_,
+                                               arg.gamma_grid_desc_m_k_,
+                                               arg.beta_grid_desc_m_k_,
+                                               arg.y_grid_desc_m_k_,
+                                               arg.numMeanVarCountIteration_,
+                                               arg.numBlockTileIteration_,
+                                               arg.kGridSize_,
+                                               arg.epsilon_,
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_mean_),
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_var_),
+                                               static_cast<int32_t*>(arg.p_workspace_count_),
+                                               arg.p_x_,
+                                               arg.p_gamma_,
+                                               arg.p_beta_,
+                                               arg.p_y_,
+                                               arg.y_elementwise_op_);
+
+            return avg_time;
+        };
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        };
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        size_t workspace_size = 0;
+
+        int welford_size = pArg_->MRaw_ * pArg_->kGridSize_;
+
+        // workspace for welford intermediate mean
+        workspace_size += welford_size * sizeof(MeanVarDataType) + 64;
+
+        // workspace for welford intermediate variance
+        workspace_size += welford_size * sizeof(MeanVarDataType) + 64;
+
+        // workspace for welford intermediate count
+        workspace_size += pArg_->kGridSize_ * sizeof(int32_t) + 64;
+
+        return (workspace_size);
+    };
+
+    void SetWorkSpacePointer(BaseArgument* pArg, void* p_workspace) const override
+    {
+        Argument* pArg_ = dynamic_cast<Argument*>(pArg);
+
+        pArg_->p_workspace_ = p_workspace;
+
+        int welford_size = pArg_->MRaw_ * pArg_->kGridSize_;
+
+        // setup buffer used for intermediate welford mean
+        pArg_->p_workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
+
+        index_t mean_space_sz = welford_size * sizeof(MeanVarDataType);
+        mean_space_sz         = math::integer_least_multiple(mean_space_sz, 64);
+
+        // setup buffer used for intermediate welford varirance
+        pArg_->p_workspace_var_ = reinterpret_cast<char*>(pArg_->p_workspace_mean_) + mean_space_sz;
+
+        index_t variance_space_sz = welford_size * sizeof(MeanVarDataType);
+        variance_space_sz         = math::integer_least_multiple(variance_space_sz, 64);
+
+        // setup buffer used for intermediate welford count
+        pArg_->p_workspace_count_ =
+            reinterpret_cast<char*>(pArg_->p_workspace_var_) + variance_space_sz;
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* p_arg_ = dynamic_cast<const Argument*>(p_arg);
+
+        constexpr index_t NumInvariantDim = Rank - NumReduceDim;
+
+        if constexpr(XYVectorDim == 0)
+        {
+            if constexpr(NumInvariantDim == 0)
+            {
+                return false;
+            }
+            else
+            {
+                if(p_arg_->xStrides_[NumInvariantDim - 1] != 1)
+                    return false;
+
+                if(p_arg_->invariant_lowest_length % XSrcVectorSize != 0)
+                    return false;
+
+                if(p_arg_->invariant_lowest_length % YDstVectorSize != 0)
+                    return false;
+            };
+        }
+        else
+        {
+            if(p_arg_->xStrides_[Rank - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % XSrcVectorSize != 0)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % YDstVectorSize != 0)
+                return false;
+        };
+
+        // if fastest dim is not reduced
+        if constexpr(GammaSrcVectorDim == 0)
+        {
+            if(p_arg_->gammaStrides_[NumInvariantDim - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % GammaSrcVectorSize != 0)
+                return false;
+        }
+        else // if fastest dim is reduced
+        {
+            if(p_arg_->gammaStrides_[Rank - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % GammaSrcVectorSize != 0)
+                return false;
+        }
+
+        // if fastest dim is not reduced
+        if constexpr(BetaSrcVectorDim == 0)
+        {
+            if(p_arg_->betaStrides_[NumInvariantDim - 1] != 1)
+                return false;
+
+            if(p_arg_->invariant_lowest_length % BetaSrcVectorSize != 0)
+                return false;
+        }
+        else // if fastest dim is reduced
+        {
+            if(p_arg_->betaStrides_[Rank - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % BetaSrcVectorSize != 0)
+                return false;
+        }
+
+        if(p_arg_->kGridSize_ <= 1)
+            return false;
+
+        return true;
+    };
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const std::vector<index_t> lengths,
+                        const std::vector<index_t> xStrides,
+                        const std::vector<index_t> gammaStrides,
+                        const std::vector<index_t> betaStrides,
+                        const std::vector<index_t> yStrides,
+                        const std::vector<index_t> reduceDims,
+                        double epsilon,
+                        const void* p_x,
+                        const void* p_gamma,
+                        const void* p_beta,
+                        void* p_y,
+                        void* p_saveMean,
+                        void* p_saveInvVar,
+                        YElementwiseOperation y_elementwise_op) override
+    {
+        // TODO
+        // Optional cache of the intermediate results (mean and InvVariance) during the
+        // forward pass could speedup in the backward
+        ignore = p_saveMean;
+        ignore = p_saveInvVar;
+
+        return std::make_unique<Argument>(lengths,
+                                          xStrides,
+                                          gammaStrides,
+                                          betaStrides,
+                                          yStrides,
+                                          reduceDims,
+                                          y_elementwise_op,
+                                          epsilon,
+                                          static_cast<const XDataType*>(p_x),
+                                          static_cast<const GammaDataType*>(p_gamma),
+                                          static_cast<const BetaDataType*>(p_beta),
+                                          static_cast<YDataType*>(p_y));
+    };
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>();
+    };
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceNormalizationSplitKImpl<" << BlockSize << ",";
+        str << "Cluster_MK_" << MThreadClusterSize << "_" << KThreadClusterSize << ",";
+        str << "Slice_MK_" << MThreadSliceSize << "_" << KThreadSliceSize << ",";
+        str << "XYSrcVectorDim_" << XYVectorDim  << ",";
+        str << "VectorSize_X" << XSrcVectorSize << "_Gamma" << GammaSrcVectorSize << "_Beta" << BetaSrcVectorSize << "_Y" << YDstVectorSize << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp
index 7b96373c0ff1fc86a79c4248db35dd5812f2c5b1..17dab08332166e23a93d5c8f25ecfec7556519ac 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
index bfde40cda2a5369015b08de568f72c786295848f..3f27c629dd6ac88a41b0072a1f51afc3b3fb3b25 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -9,7 +9,7 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
-#include "ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
@@ -20,16 +20,18 @@ namespace device {
 
 template <typename InDataType,
           typename OutDataType,
-          typename AccDataType,
+          typename IndexDataType, // enable if OutputIndex == true
+          typename ComputeDataType,
           ck::ReduceTensorOp ReduceOpId,
-          bool OuputIndex,
+          bool OutputIndex,
           ck::index_t BlockSize,
           ck::index_t ReduceMThreadClusterSize,
           ck::index_t ReduceKThreadClusterSize,
           ck::index_t ReduceMThreadSliceSize,
           ck::index_t ReduceKThreadSliceSize,
           ck::index_t InSrcOutDstVectorSize>
-struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd<ReduceOpId>
+struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C
+    : public DevicePoolFwd<4, 2, InDataType, OutDataType, IndexDataType, ReduceOpId, OutputIndex>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
@@ -38,7 +40,8 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
     static constexpr auto I4 = Number<4>{};
     static constexpr auto I5 = Number<5>{};
 
-    using IndexDataType = int32_t;
+    static constexpr index_t InOutRank  = 4;
+    static constexpr index_t WindowRank = 2;
 
     using ReduceOperation = typename reduce_binary_operator<ReduceOpId>::opType;
 
@@ -59,12 +62,12 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
 
     static auto MakeABGridDescriptor_A_M_K_B_M(ck::index_t N,
                                                ck::index_t C,
-                                               std::array<ck::index_t, 2> input_spatial_lengths,
-                                               std::array<ck::index_t, 2> window_spatial_lengths,
-                                               std::array<ck::index_t, 2> output_spatial_lengths,
-                                               std::array<ck::index_t, 2> window_strides,
-                                               std::array<ck::index_t, 2> input_left_pads,
-                                               std::array<ck::index_t, 2> input_right_pads)
+                                               std::vector<ck::index_t> input_spatial_lengths,
+                                               std::vector<ck::index_t> window_spatial_lengths,
+                                               std::vector<ck::index_t> output_spatial_lengths,
+                                               std::vector<ck::index_t> window_strides,
+                                               std::vector<ck::index_t> input_left_pads,
+                                               std::vector<ck::index_t> input_right_pads)
     {
         const index_t Hi = input_spatial_lengths[0];
         const index_t Wi = input_spatial_lengths[1];
@@ -141,9 +144,7 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
         return make_tuple(in_grid_desc_reducem_reducek, out_grid_desc_reducem);
     }
 
-    using ABGridDescs = decltype(
-        MakeABGridDescriptor_A_M_K_B_M(1, 1, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}));
-
+    using ABGridDescs   = decltype(MakeABGridDescriptor_A_M_K_B_M(1, 1, {}, {}, {}, {}, {}, {}));
     using AGridDesc_M_K = remove_cvref_t<decltype(ABGridDescs{}[I0])>;
     using BGridDesc_M   = remove_cvref_t<decltype(ABGridDescs{}[I1])>;
 
@@ -152,15 +153,15 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
     {
         Argument(const InDataType* p_in_dev,
                  OutDataType* p_out_dev,
-                 int* p_out_indices_dev,
+                 IndexDataType* p_out_indices_dev,
                  ck::index_t N,
                  ck::index_t C,
-                 std::array<ck::index_t, 2>& input_spatial_lengths,
-                 std::array<ck::index_t, 2>& window_spatial_lengths,
-                 std::array<ck::index_t, 2>& output_spatial_lengths,
-                 std::array<ck::index_t, 2>& window_strides,
-                 std::array<ck::index_t, 2>& input_left_pads,
-                 std::array<ck::index_t, 2>& input_right_pads)
+                 std::vector<ck::index_t>& input_spatial_lengths,
+                 std::vector<ck::index_t>& window_spatial_lengths,
+                 std::vector<ck::index_t>& output_spatial_lengths,
+                 std::vector<ck::index_t>& window_strides,
+                 std::vector<ck::index_t>& input_left_pads,
+                 std::vector<ck::index_t>& input_right_pads)
             : p_in_dev_{p_in_dev},
               p_out_dev_{p_out_dev},
               p_out_indices_dev_{p_out_indices_dev},
@@ -190,7 +191,7 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
 
         const InDataType* p_in_dev_;
         OutDataType* p_out_dev_;
-        int* p_out_indices_dev_;
+        IndexDataType* p_out_indices_dev_;
         AGridDesc_M_K a_grid_desc_m_k_;
         BGridDesc_M b_grid_desc_m_;
         InElementwiseOperation in_element_op_;
@@ -208,7 +209,7 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
             using gridwise_reduce =
                 GridwiseReduction_mk_to_m_threadwise<InDataType,
                                                      OutDataType,
-                                                     AccDataType,
+                                                     ComputeDataType,
                                                      IndexDataType,
                                                      AGridDesc_M_K,
                                                      BGridDesc_M,
@@ -224,17 +225,19 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
                                                      InSrcOutDstVectorSize,
                                                      InSrcOutDstVectorSize>;
 
-            const auto kernel = kernel_reduce_threadwise<gridwise_reduce,
-                                                         OuputIndex,
-                                                         false, // don't have index input
-                                                         InDataType,
-                                                         OutDataType,
-                                                         AccDataType,
-                                                         IndexDataType,
-                                                         AGridDesc_M_K,
-                                                         BGridDesc_M,
-                                                         InElementwiseOperation,
-                                                         AccElementwiseOperation>;
+            const auto kernel =
+                kernel_reduce_threadwise<gridwise_reduce,
+                                         OutputIndex,
+                                         true,  // pooling need to return global index
+                                         false, // don't have index input
+                                         InDataType,
+                                         OutDataType,
+                                         ComputeDataType,
+                                         IndexDataType,
+                                         AGridDesc_M_K,
+                                         BGridDesc_M,
+                                         InElementwiseOperation,
+                                         AccElementwiseOperation>;
 
             ck::index_t ReduceM = arg.a_grid_desc_m_k_.GetLength(I0);
 
@@ -280,22 +283,42 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
     MakeArgumentPointer(const void* p_in_dev,
                         void* p_out_dev,
                         void* p_out_indices_dev,
-                        ck::index_t N,
-                        ck::index_t C,
-                        std::array<ck::index_t, 2> input_spatial_lengths,
-                        std::array<ck::index_t, 2> window_spatial_lengths,
-                        std::array<ck::index_t, 2> output_spatial_lengths,
-                        std::array<ck::index_t, 2> window_strides,
-                        std::array<ck::index_t, 2> input_left_pads,
-                        std::array<ck::index_t, 2> input_right_pads) override
+                        std::vector<ck::index_t> input_lengths,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> output_lengths,
+                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads,
+                        std::vector<ck::index_t> pooling_dims) override
     {
+        if(input_lengths.size() != InOutRank || window_lengths.size() != WindowRank ||
+           input_lengths.size() != InOutRank || window_strides.size() != WindowRank ||
+           input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+            throw std::runtime_error("dimension is incorrect");
+
+        if(pooling_dims != std::vector<ck::index_t>{2, 3})
+            throw std::runtime_error("pooling_dims only support {2, 3} in pool2d so far");
+
+        index_t N  = input_lengths[0];
+        index_t C  = input_lengths[1];
+        index_t Hi = input_lengths[2];
+        index_t Wi = input_lengths[3];
+        index_t Ho = output_lengths[2];
+        index_t Wo = output_lengths[3];
+
+        std::vector<ck::index_t> input_spatial_lengths  = {Hi, Wi};
+        std::vector<ck::index_t> output_spatial_lengths = {Ho, Wo};
+
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_dev),
                                           static_cast<OutDataType*>(p_out_dev),
-                                          static_cast<int*>(p_out_indices_dev),
+                                          static_cast<IndexDataType*>(p_out_indices_dev),
                                           N,
                                           C,
                                           input_spatial_lengths,
-                                          window_spatial_lengths,
+                                          window_lengths,
                                           output_spatial_lengths,
                                           window_strides,
                                           input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0ab6c247584881bd72723b6cda2f412aea9a2e1c
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
@@ -0,0 +1,357 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType, // enable if OutputIndex == true
+          typename ComputeDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool OutputIndex,
+          ck::index_t BlockSize,
+          ck::index_t MThreadClusterSize,
+          ck::index_t KThreadClusterSize,
+          ck::index_t MThreadSliceSize,
+          ck::index_t KThreadSliceSize,
+          ck::index_t InSrcOutDstVectorSize>
+struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
+    : public DevicePoolFwd<5, 3, InDataType, OutDataType, IndexDataType, ReduceOpId, OutputIndex>
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr index_t InOutRank  = 5;
+    static constexpr index_t WindowRank = 3;
+
+    using ReduceOperation = typename reduce_binary_operator<ReduceOpId>::opType;
+
+    using InElementwiseOperation =
+        typename reduce_unary_operator<ReduceOpId, true, true>::InElementwiseOperation;
+
+    using AccElementwiseOperation =
+        typename reduce_unary_operator<ReduceOpId, true, true>::AccElementwiseOperation;
+
+    // for NDHWC, the dim C is the vector Dim for both input and output in memory, which is not
+    // reduced.
+    static constexpr index_t InSrcOutDstVectorDim = 0;
+
+    static constexpr ck::index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr ck::index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static auto MakeABGridDescriptor_A_M_K_B_M(ck::index_t N,
+                                               ck::index_t C,
+                                               std::vector<ck::index_t> input_spatial_lengths,
+                                               std::vector<ck::index_t> window_spatial_lengths,
+                                               std::vector<ck::index_t> output_spatial_lengths,
+                                               std::vector<ck::index_t> window_strides,
+                                               std::vector<ck::index_t> input_left_pads,
+                                               std::vector<ck::index_t> input_right_pads)
+    {
+        const index_t Di = input_spatial_lengths[0];
+        const index_t Hi = input_spatial_lengths[1];
+        const index_t Wi = input_spatial_lengths[2];
+
+        const index_t Do = output_spatial_lengths[0];
+        const index_t Ho = output_spatial_lengths[1];
+        const index_t Wo = output_spatial_lengths[2];
+
+        const index_t Z = window_spatial_lengths[0];
+        const index_t Y = window_spatial_lengths[1];
+        const index_t X = window_spatial_lengths[2];
+
+        const index_t ConvStrideD = window_strides[0];
+        const index_t ConvStrideH = window_strides[1];
+        const index_t ConvStrideW = window_strides[2];
+
+        const index_t InLeftPadD = input_left_pads[0];
+        const index_t InLeftPadH = input_left_pads[1];
+        const index_t InLeftPadW = input_left_pads[2];
+
+        const index_t InRightPadD = input_right_pads[0];
+        const index_t InRightPadH = input_right_pads[1];
+        const index_t InRightPadW = input_right_pads[2];
+
+        const index_t MRaw = N * Do * Ho * Wo * C;
+        const index_t MPad = math::integer_least_multiple(MRaw, M_BlockTileSize) - MRaw;
+
+        const index_t KRaw = Z * Y * X;
+        const index_t KPad = math::integer_least_multiple(KRaw, K_BlockTileSize) - KRaw;
+
+        // A[ReduceM, ReduceK]
+        const auto in_grid_desc_n_di_hi_wi_c =
+            make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
+
+        const auto in_grid_desc_n_dip_hip_wip_c = transform_tensor_descriptor(
+            in_grid_desc_n_di_hi_wi_c,
+            make_tuple(make_pass_through_transform(N),
+                       make_pad_transform(Di, InLeftPadD, InRightPadD),
+                       make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                       make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                       make_pass_through_transform(C)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+        const auto in_grid_desc_n_z_do_y_ho_x_wo_c = transform_tensor_descriptor(
+            in_grid_desc_n_dip_hip_wip_c,
+            make_tuple(make_pass_through_transform(N),
+                       make_embed_transform(make_tuple(Z, Do), make_tuple(I1, ConvStrideD)),
+                       make_embed_transform(make_tuple(Y, Ho), make_tuple(I1, ConvStrideH)),
+                       make_embed_transform(make_tuple(X, Wo), make_tuple(I1, ConvStrideW)),
+                       make_pass_through_transform(C)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+            make_tuple(Sequence<0>{},
+                       Sequence<1, 2>{},
+                       Sequence<3, 4>{},
+                       Sequence<5, 6>{},
+                       Sequence<7>{}));
+
+        const auto in_grid_desc_reducemraw_reducekraw = transform_tensor_descriptor(
+            in_grid_desc_n_z_do_y_ho_x_wo_c,
+            make_tuple(make_merge_transform(make_tuple(N, Do, Ho, Wo, C)),
+                       make_merge_transform(make_tuple(Z, Y, X))),
+            make_tuple(Sequence<0, 2, 4, 6, 7>{}, Sequence<1, 3, 5>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        const auto in_grid_desc_reducem_reducek = transform_tensor_descriptor(
+            in_grid_desc_reducemraw_reducekraw,
+            make_tuple(make_right_pad_transform(MRaw, MPad), make_right_pad_transform(KRaw, KPad)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        // B[ReduceM]
+        const auto out_grid_desc_reducemraw =
+            make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo * C));
+
+        const auto out_grid_desc_reducem =
+            transform_tensor_descriptor(out_grid_desc_reducemraw,
+                                        make_tuple(make_right_pad_transform(MRaw, MPad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+
+        return make_tuple(in_grid_desc_reducem_reducek, out_grid_desc_reducem);
+    }
+
+    using ABGridDescs   = decltype(MakeABGridDescriptor_A_M_K_B_M(1, 1, {}, {}, {}, {}, {}, {}));
+    using AGridDesc_M_K = remove_cvref_t<decltype(ABGridDescs{}[I0])>;
+    using BGridDesc_M   = remove_cvref_t<decltype(ABGridDescs{}[I1])>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const InDataType* p_in_dev,
+                 OutDataType* p_out_dev,
+                 IndexDataType* p_out_indices_dev,
+                 ck::index_t N,
+                 ck::index_t C,
+                 std::vector<ck::index_t>& input_spatial_lengths,
+                 std::vector<ck::index_t>& window_spatial_lengths,
+                 std::vector<ck::index_t>& output_spatial_lengths,
+                 std::vector<ck::index_t>& window_strides,
+                 std::vector<ck::index_t>& input_left_pads,
+                 std::vector<ck::index_t>& input_right_pads)
+            : p_in_dev_{p_in_dev},
+              p_out_dev_{p_out_dev},
+              p_out_indices_dev_{p_out_indices_dev},
+              a_grid_desc_m_k_{},
+              b_grid_desc_m_{}
+        {
+            const auto descs = MakeABGridDescriptor_A_M_K_B_M(N,
+                                                              C,
+                                                              input_spatial_lengths,
+                                                              window_spatial_lengths,
+                                                              output_spatial_lengths,
+                                                              window_strides,
+                                                              input_left_pads,
+                                                              input_right_pads);
+
+            a_grid_desc_m_k_ = descs[I0];
+            b_grid_desc_m_   = descs[I1];
+
+            invariant_lowest_length_ = C;
+
+            int32_t reduceLength =
+                window_spatial_lengths[0] * window_spatial_lengths[1] * window_spatial_lengths[2];
+
+            std::tie(in_element_op_, acc_element_op_) =
+                reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(reduceLength);
+        }
+
+        const InDataType* p_in_dev_;
+        OutDataType* p_out_dev_;
+        IndexDataType* p_out_indices_dev_;
+        AGridDesc_M_K a_grid_desc_m_k_;
+        BGridDesc_M b_grid_desc_m_;
+        InElementwiseOperation in_element_op_;
+        AccElementwiseOperation acc_element_op_;
+
+        // for checking vector load/store
+        ck::index_t invariant_lowest_length_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            using gridwise_reduce =
+                GridwiseReduction_mk_to_m_threadwise<InDataType,
+                                                     OutDataType,
+                                                     ComputeDataType,
+                                                     IndexDataType,
+                                                     AGridDesc_M_K,
+                                                     BGridDesc_M,
+                                                     ReduceOperation,
+                                                     InElementwiseOperation,
+                                                     AccElementwiseOperation,
+                                                     InMemoryDataOperationEnum::Set,
+                                                     false, // propagate_nan
+                                                     BlockSize,
+                                                     MThreadSliceSize,
+                                                     KThreadSliceSize,
+                                                     InSrcOutDstVectorDim,
+                                                     InSrcOutDstVectorSize,
+                                                     InSrcOutDstVectorSize>;
+
+            const auto kernel =
+                kernel_reduce_threadwise<gridwise_reduce,
+                                         OutputIndex,
+                                         true,  // pooling need to return global index
+                                         false, // don't have index input
+                                         InDataType,
+                                         OutDataType,
+                                         ComputeDataType,
+                                         IndexDataType,
+                                         AGridDesc_M_K,
+                                         BGridDesc_M,
+                                         InElementwiseOperation,
+                                         AccElementwiseOperation>;
+
+            ck::index_t M = arg.a_grid_desc_m_k_.GetLength(I0);
+
+            const index_t grid_size = (M / M_BlockTileSize);
+
+            return launch_and_time_kernel(stream_config,
+                                          kernel,
+                                          dim3(grid_size),
+                                          dim3(BlockSize),
+                                          0,
+                                          arg.a_grid_desc_m_k_,
+                                          arg.b_grid_desc_m_,
+                                          arg.in_element_op_,
+                                          arg.acc_element_op_,
+                                          float(1),
+                                          arg.p_in_dev_,
+                                          nullptr,
+                                          float(0),
+                                          arg.p_out_dev_,
+                                          arg.p_out_indices_dev_);
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
+
+        if(pArg->invariant_lowest_length_ % InSrcOutDstVectorSize != 0)
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in_dev,
+                        void* p_out_dev,
+                        void* p_out_indices_dev,
+                        std::vector<ck::index_t> input_lengths,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> output_lengths,
+                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads,
+                        std::vector<ck::index_t> pooling_dims) override
+    {
+        if(input_lengths.size() != InOutRank || window_lengths.size() != WindowRank ||
+           input_lengths.size() != InOutRank || window_strides.size() != WindowRank ||
+           input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+            throw std::runtime_error("dimension is incorrect");
+
+        if(pooling_dims != std::vector<ck::index_t>{2, 3, 4})
+            throw std::runtime_error("pooling_dims only support {2, 3, 4} in pool3d so far");
+
+        index_t N  = input_lengths[0];
+        index_t C  = input_lengths[1];
+        index_t Di = input_lengths[2];
+        index_t Hi = input_lengths[3];
+        index_t Wi = input_lengths[4];
+        index_t Do = output_lengths[2];
+        index_t Ho = output_lengths[3];
+        index_t Wo = output_lengths[4];
+
+        std::vector<ck::index_t> input_spatial_lengths  = {Di, Hi, Wi};
+        std::vector<ck::index_t> output_spatial_lengths = {Do, Ho, Wo};
+
+        return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_dev),
+                                          static_cast<OutDataType*>(p_out_dev),
+                                          static_cast<IndexDataType*>(p_out_indices_dev),
+                                          N,
+                                          C,
+                                          input_spatial_lengths,
+                                          window_lengths,
+                                          output_spatial_lengths,
+                                          window_strides,
+                                          input_left_pads,
+                                          input_right_pads);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<" << BlockSize << ",";
+        str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
+        str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
+        str <<"InSrcOutDstVectorSize_" << InSrcOutDstVectorSize << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..7334da0e35af5ce92665decc4897ebe48393b9c3
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/device_put_element.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// output[indices] = input
+template <typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation,
+          InMemoryDataOperationEnum MemOp,
+          ck::index_t InVectorSize>
+struct DevicePutElementImpl
+    : public DevicePutElement<InDataType, IndexDataType, OutDataType, ElementwiseOperation, MemOp>
+{
+    template <typename Desc_M>
+    static auto PadDescriptor_M_1d(Desc_M desc_m, index_t gridSize, index_t blockSize)
+    {
+        constexpr auto I0 = Number<0>{};
+
+        const auto m            = desc_m.GetLength(I0);
+        const index_t loop_step = gridSize * blockSize * InVectorSize;
+        const auto pad          = math::integer_least_multiple(m, loop_step) - m;
+        const auto desc_m_pad =
+            transform_tensor_descriptor(desc_m,
+                                        make_tuple(make_right_pad_transform(m, pad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+        return desc_m_pad;
+    }
+
+    static auto MakeDescriptor_M(index_t length, index_t gridSize, index_t blockSize)
+    {
+        const auto desc_m = make_naive_tensor_descriptor_packed(make_tuple(length));
+        return PadDescriptor_M_1d(desc_m, gridSize, blockSize);
+    }
+
+    using InGrid1dDesc = decltype(MakeDescriptor_M(1, 1, 1));
+
+    using GridwisePutElement = GridwisePutElement_1D<InGrid1dDesc,
+                                                     InDataType,
+                                                     IndexDataType,
+                                                     OutDataType,
+                                                     ElementwiseOperation,
+                                                     MemOp,
+                                                     InVectorSize>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const InDataType* p_input,
+                 const IndexDataType* p_indices,
+                 OutDataType* p_output,
+                 index_t input_length,
+                 ElementwiseOperation elementwise_op)
+            : p_input_{p_input},
+              p_indices_{p_indices},
+              p_output_{p_output},
+              input_length_raw_{input_length},
+              elementwise_op_{elementwise_op},
+              blockSize_{256}
+        {
+        }
+
+        const InDataType* p_input_;
+        const IndexDataType* p_indices_;
+        OutDataType* p_output_;
+        index_t input_length_raw_;
+        ElementwiseOperation elementwise_op_;
+        index_t blockSize_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            index_t gridSize = getAvailableComputeUnitCount(stream_config);
+            InGrid1dDesc in_grid_desc =
+                MakeDescriptor_M(arg.input_length_raw_, gridSize, arg.blockSize_);
+
+            const auto kernel = kernel_put_element_1d<GridwisePutElement,
+                                                      InGrid1dDesc,
+                                                      InDataType,
+                                                      IndexDataType,
+                                                      OutDataType,
+                                                      ElementwiseOperation>;
+
+            float elapsed_time = launch_and_time_kernel(stream_config,
+                                                        kernel,
+                                                        dim3(gridSize),
+                                                        dim3(arg.blockSize_),
+                                                        0,
+                                                        in_grid_desc,
+                                                        arg.p_input_,
+                                                        arg.p_indices_,
+                                                        arg.p_output_,
+                                                        arg.elementwise_op_);
+            return elapsed_time;
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
+
+        if(pArg->input_length_raw_ % InVectorSize != 0)
+        {
+            return false;
+        }
+        return true;
+    }
+
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_input,
+                                                      const void* p_indices,
+                                                      void* p_output,
+                                                      index_t input_length,
+                                                      index_t,
+                                                      ElementwiseOperation elementwise_op) override
+    {
+        return std::make_unique<Argument>(static_cast<const InDataType*>(p_input),
+                                          static_cast<const IndexDataType*>(p_indices),
+                                          static_cast<OutDataType*>(p_output),
+                                          input_length,
+                                          elementwise_op);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp b/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp
index 5dc051be3cb7f23fef076921e826a7e55e017d76..2481c5c76971db6bb2ddc390510089af5f656ae6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp b/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp
index 93855eb33e1ba2a35e8fb31a9cc6994338d26aab..bf3deeb57acbdff95995f5da096ab574140bcb42 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -40,8 +40,16 @@ template <typename InDataType,
           index_t InSrcVectorDim,
           index_t InSrcVectorSize,
           index_t OutDstVectorSize>
-struct DeviceReduceMultiBlock
-    : public DeviceReduce<Rank, NumReduceDim, InElementwiseOperation, AccElementwiseOperation>
+struct DeviceReduceMultiBlock : public DeviceReduce<InDataType,
+                                                    AccDataType,
+                                                    OutDataType,
+                                                    Rank,
+                                                    NumReduceDim,
+                                                    ReduceOperation,
+                                                    InElementwiseOperation,
+                                                    AccElementwiseOperation,
+                                                    PropagateNan,
+                                                    OutputIndex>
 {
     static_assert(Rank <= 6, "Bigger Rank size is not supported!");
     static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize,
@@ -67,8 +75,8 @@ struct DeviceReduceMultiBlock
     static constexpr bool use_multiblock =
         (OutMemoryDataOperation == InMemoryDataOperationEnum::AtomicAdd);
 
-    static_assert(ck::reduce::InMemoryDataOperatonSupportedOnDataType<OutMemoryDataOperation,
-                                                                      OutDataType>::value,
+    static_assert(ck::reduce::InMemoryDataOperationSupportedOnDataType<OutMemoryDataOperation,
+                                                                       OutDataType>::value,
                   "The OutDataType must support the specified OutMemoryDataOperation!");
 
     static_assert(!use_multiblock || (use_multiblock && !OutputIndex),
@@ -209,8 +217,8 @@ struct DeviceReduceMultiBlock
                  const std::array<index_t, NumDstDim> outLengths,
                  const std::array<index_t, NumDstDim> outStrides,
                  const std::array<int, NumReduceDim> reduceDims,
-                 float alpha,
-                 float beta,
+                 double alpha,
+                 double beta,
                  const InDataType* in_dev,
                  const IndexDataType* in_index_dev,
                  OutDataType* out_dev,
@@ -494,8 +502,8 @@ struct DeviceReduceMultiBlock
                         const std::array<index_t, NumDstDim> outLengths,
                         const std::array<index_t, NumDstDim> outStrides,
                         const std::array<int, NumReduceDim> reduceDims,
-                        float alpha,
-                        float beta,
+                        double alpha,
+                        double beta,
                         const void* in_dev,
                         const void* in_index_dev,
                         void* out_dev,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp b/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
index 05e14f080efe44d1715acc9a506e4c4bf3c4a9bd..6c5895b010659cafb98e4a78bc6587d0222e02bd 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -28,6 +28,7 @@ template <typename InDataType,
           typename AccElementwiseOperation,
           bool PropagateNan,
           bool OutputIndex,
+          bool TransformIndexKtoGlobal,
           bool HaveIndexInputIfOutputIndex,
           index_t BlockSize,
           index_t MThreadSliceSize,
@@ -35,8 +36,17 @@ template <typename InDataType,
           index_t InSrcVectorDim,
           index_t InSrcVectorSize,
           index_t OutDstVectorSize>
-struct DeviceReduceThreadWise
-    : public DeviceReduce<Rank, NumReduceDim, InElementwiseOperation, AccElementwiseOperation>
+struct DeviceReduceThreadWise : public DeviceReduce<InDataType,
+                                                    AccDataType,
+                                                    OutDataType,
+                                                    Rank,
+                                                    NumReduceDim,
+                                                    ReduceOperation,
+                                                    InElementwiseOperation,
+                                                    AccElementwiseOperation,
+                                                    PropagateNan,
+                                                    OutputIndex>
+
 {
     static_assert(Rank <= 6, "Bigger Rank size is not supported!");
 
@@ -156,8 +166,8 @@ struct DeviceReduceThreadWise
                  const std::array<index_t, NumDstDim> outLengths,
                  const std::array<index_t, NumDstDim> outStrides,
                  const std::array<int, NumReduceDim> reduceDims,
-                 float alpha,
-                 float beta,
+                 double alpha,
+                 double beta,
                  const InDataType* in_dev,
                  OutDataType* out_dev,
                  IndexDataType* out_index_dev,
@@ -251,6 +261,7 @@ struct DeviceReduceThreadWise
 
             const auto kernel = kernel_reduce_threadwise<GridwiseReduce,
                                                          OutputIndex,
+                                                         TransformIndexKtoGlobal,
                                                          HaveIndexInput,
                                                          InDataType,
                                                          OutDataType,
@@ -332,8 +343,8 @@ struct DeviceReduceThreadWise
                         const std::array<index_t, NumDstDim> outLengths,
                         const std::array<index_t, NumDstDim> outStrides,
                         const std::array<int, NumReduceDim> reduceDims,
-                        float alpha,
-                        float beta,
+                        double alpha,
+                        double beta,
                         const void* in_dev,
                         const void* in_index_dev,
                         void* out_dev,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
index 8630a2c6e227c8b929a3644e75af8f849d32e92c..8eff9d241509bdc4b1cc3993a39f0e3c7fa6828d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -38,16 +38,9 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
                                                 OutDataType,
                                                 InElementwiseOp,
                                                 AccElementwiseOp,
-                                                Rank>
+                                                Rank,
+                                                NumReduceDim>
 {
-    static constexpr index_t kRank            = Rank;
-    static constexpr index_t kNumReduceDim    = NumReduceDim;
-    static constexpr index_t kNumInvariantDim = Rank - NumReduceDim;
-
-    virtual index_t GetRank() const override { return kRank; }
-
-    virtual index_t GetNumReduceDim() const override { return kNumReduceDim; }
-
     static constexpr index_t NumInvariantDim = Rank - NumReduceDim;
 
     static constexpr index_t NumSrcDim = Rank;
@@ -156,19 +149,20 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
         Argument(const std::vector<index_t> inLengths,
                  const std::vector<index_t> inStrides,
                  const std::vector<index_t> reduceDims,
-                 AccDataType alpha,
-                 AccDataType beta,
+                 double alpha,
+                 double beta,
                  const InDataType* in_dev,
                  OutDataType* out_dev,
                  InElementwiseOp in_elementwise_op,
                  AccElementwiseOp acc_elementwise_op)
-            : alpha_{alpha},
-              beta_{beta},
-              in_dev_{in_dev},
+            : in_dev_{in_dev},
               out_dev_{out_dev},
               in_elementwise_op_{in_elementwise_op},
               acc_elementwise_op_{acc_elementwise_op}
         {
+            alpha_ = static_cast<AccDataType>(alpha);
+            beta_  = static_cast<AccDataType>(beta);
+
             if(Rank != inLengths.size() || Rank != inStrides.size() ||
                NumReduceDim != reduceDims.size())
             {
@@ -286,13 +280,13 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
     {
         if constexpr(InSrcVectorDim == 0)
         {
-            if constexpr(kNumInvariantDim == 0)
+            if constexpr(NumInvariantDim == 0)
             {
                 return false;
             }
             else
             {
-                if(arg.inStrides_[kNumInvariantDim - 1] != 1 && InSrcVectorSize != 1)
+                if(arg.inStrides_[NumInvariantDim - 1] != 1 && InSrcVectorSize != 1)
                 {
                     return false;
                 }
@@ -315,7 +309,7 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
         }
 
         // To improve
-        if(kNumInvariantDim > 0 && arg.invariant_lowest_length_ % OutDstVectorSize != 0)
+        if(NumInvariantDim > 0 && arg.invariant_lowest_length_ % OutDstVectorSize != 0)
         {
             return false;
         }
@@ -336,8 +330,8 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
     static auto MakeArgument(const std::vector<index_t> inLengths,
                              const std::vector<index_t> inStrides,
                              const std::vector<int> reduceDims,
-                             const AccDataType alpha,
-                             const AccDataType beta,
+                             double alpha,
+                             double beta,
                              const InDataType* in_dev,
                              OutDataType* out_dev,
                              InElementwiseOp in_elementwise_op,
@@ -375,8 +369,8 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
     std::unique_ptr<BaseArgument> MakeArgumentPointer(const std::vector<index_t> inLengths,
                                                       const std::vector<index_t> inStrides,
                                                       const std::vector<int> reduceDims,
-                                                      const void* alpha,
-                                                      const void* beta,
+                                                      double alpha,
+                                                      double beta,
                                                       const void* in_dev,
                                                       void* out_dev,
                                                       InElementwiseOp in_elementwise_op,
@@ -385,8 +379,8 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
         return std::make_unique<Argument>(inLengths,
                                           inStrides,
                                           reduceDims,
-                                          *static_cast<const AccDataType*>(alpha),
-                                          *static_cast<const AccDataType*>(beta),
+                                          alpha,
+                                          beta,
                                           static_cast<const InDataType*>(in_dev),
                                           static_cast<OutDataType*>(out_dev),
                                           in_elementwise_op,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_sparse_embedding3_forward_layernorm.hpp b/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
similarity index 60%
rename from include/ck/tensor_operation/gpu/device/impl/device_sparse_embedding3_forward_layernorm.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
index 1f2b46edd3c1f70edbd6247c1919fad9e0d6b1ad..7a62ec04650a94d1b9f458e425ddf3b14881091d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_sparse_embedding3_forward_layernorm.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -12,7 +12,7 @@
 #include "ck/utility/common_header.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_sparse_embedding3_forward_layernorm.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -24,16 +24,17 @@ template <typename EmbType,
           typename BetaDataType,
           typename AccDataType,
           typename OutType,
+          typename EmbElementwiseOperation,
           ck::index_t BlockSize,
           ck::index_t DimClusterSize,
           ck::index_t RowClusterSize,
           ck::index_t DimPerBlock,
           ck::index_t RowPerBlock,
           ck::index_t DimThreadSize,
-          ck::index_t RowVectorSize>
-struct DeviceSparseEmbedding3ForwardLayernorm : public BaseOperator
+          ck::index_t RowVectorSize,
+          ck::index_t NumEmbeddings>
+struct DeviceSparseEmbeddingsForwardLayernorm : public BaseOperator
 {
-
     static auto MakeOutputDescriptor(const index_t index_length, const index_t rows)
     {
         return make_naive_tensor_descriptor_packed(make_tuple(index_length, rows));
@@ -42,96 +43,79 @@ struct DeviceSparseEmbedding3ForwardLayernorm : public BaseOperator
     struct Argument : public BaseArgument
     {
         Argument(OutType* p_out,
-                 const EmbType* p_emb_a,
-                 const EmbType* p_emb_b,
-                 const EmbType* p_emb_c,
-                 const IndexType* p_index_a,
-                 const IndexType* p_index_b,
-                 const IndexType* p_index_c,
+                 const ck::Array<EmbType*, NumEmbeddings>& p_embs,
+                 const ck::Array<IndexType*, NumEmbeddings>& p_indexs,
                  const GammaDataType* p_gamma,
                  const BetaDataType* p_beta,
-                 const ck::index_t NumRows,
                  const ck::index_t EmbeddingDim,
                  const ck::index_t IndexLength,
-                 const AccDataType epsilon)
+                 const AccDataType epsilon,
+                 const EmbElementwiseOperation emb_elementwise_op)
             : p_out_(p_out),
-              p_emb_a_(p_emb_a),
-              p_emb_b_(p_emb_b),
-              p_emb_c_(p_emb_c),
-              p_index_a_(p_index_a),
-              p_index_b_(p_index_b),
-              p_index_c_(p_index_c),
+              p_embs_(p_embs),
+              p_indexs_(p_indexs),
               p_gamma_(p_gamma),
               p_beta_(p_beta),
-              NumRows_(NumRows),
               EmbeddingDim_(EmbeddingDim),
               IndexLength_(IndexLength),
-              epsilon_(epsilon)
+              epsilon_(epsilon),
+              emb_elementwise_op_(emb_elementwise_op)
         {
             grid_size_ = (IndexLength + DimClusterSize - 1) / DimClusterSize;
         }
 
         OutType* p_out_;
-        const EmbType* p_emb_a_;
-        const EmbType* p_emb_b_;
-        const EmbType* p_emb_c_;
-        const IndexType* p_index_a_;
-        const IndexType* p_index_b_;
-        const IndexType* p_index_c_;
+        ck::Array<EmbType*, NumEmbeddings> p_embs_;
+        ck::Array<IndexType*, NumEmbeddings> p_indexs_;
         const GammaDataType* p_gamma_;
         const BetaDataType* p_beta_;
-        ck::index_t NumRows_;
         ck::index_t EmbeddingDim_;
         ck::index_t IndexLength_;
         AccDataType epsilon_;
+        EmbElementwiseOperation emb_elementwise_op_;
 
         size_t grid_size_;
     };
 
-    virtual std::unique_ptr<BaseArgument> MakeArgumentPointer(void* p_out,
-                                                              const void* p_emb_a,
-                                                              const void* p_emb_b,
-                                                              const void* p_emb_c,
-                                                              const void* p_index_a,
-                                                              const void* p_index_b,
-                                                              const void* p_index_c,
-                                                              const void* p_gamma,
-                                                              const void* p_beta,
-                                                              ck::index_t NumRows,
-                                                              ck::index_t EmbeddingDim,
-                                                              ck::index_t IndexLength,
-                                                              const AccDataType epsilon)
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(void* p_out,
+                        const ck::Array<EmbType*, NumEmbeddings>& p_embs,
+                        const ck::Array<IndexType*, NumEmbeddings>& p_indexs,
+                        const void* p_gamma,
+                        const void* p_beta,
+                        ck::index_t EmbeddingDim,
+                        ck::index_t IndexLength,
+                        const AccDataType epsilon,
+                        const EmbElementwiseOperation emb_elementwise_op)
     {
         return std::make_unique<Argument>(reinterpret_cast<OutType*>(p_out),
-                                          reinterpret_cast<const EmbType*>(p_emb_a),
-                                          reinterpret_cast<const EmbType*>(p_emb_b),
-                                          reinterpret_cast<const EmbType*>(p_emb_c),
-                                          reinterpret_cast<const IndexType*>(p_index_a),
-                                          reinterpret_cast<const IndexType*>(p_index_b),
-                                          reinterpret_cast<const IndexType*>(p_index_c),
+                                          p_embs,
+                                          p_indexs,
                                           reinterpret_cast<const GammaDataType*>(p_gamma),
                                           reinterpret_cast<const BetaDataType*>(p_beta),
-                                          NumRows,
                                           EmbeddingDim,
                                           IndexLength,
-                                          epsilon);
+                                          epsilon,
+                                          emb_elementwise_op);
     }
 
     using GridwiseSparseEmbedding =
-        GridwiseSparseEmbedding3ForwardLayernorm<EmbType,
+        GridwiseSparseEmbeddingsForwardLayernorm<EmbType,
                                                  IndexType,
                                                  GammaDataType,
                                                  BetaDataType,
                                                  AccDataType,
                                                  OutType,
                                                  decltype(MakeOutputDescriptor(1, 1)),
+                                                 EmbElementwiseOperation,
                                                  BlockSize,
                                                  DimClusterSize,
                                                  RowClusterSize,
                                                  DimPerBlock,
                                                  RowPerBlock,
                                                  DimThreadSize,
-                                                 RowVectorSize>;
+                                                 RowVectorSize,
+                                                 NumEmbeddings>;
 
     struct Invoker : public BaseInvoker
     {
@@ -139,14 +123,16 @@ struct DeviceSparseEmbedding3ForwardLayernorm : public BaseOperator
         {
             auto out_desc = MakeOutputDescriptor(arg.IndexLength_, arg.EmbeddingDim_);
             const auto kernel_main =
-                kernel_sparse_embedding3_forward_layernorm<GridwiseSparseEmbedding,
+                kernel_sparse_embeddings_forward_layernorm<GridwiseSparseEmbedding,
                                                            EmbType,
                                                            IndexType,
                                                            GammaDataType,
                                                            BetaDataType,
                                                            AccDataType,
                                                            OutType,
-                                                           decltype(out_desc)>;
+                                                           decltype(out_desc),
+                                                           EmbElementwiseOperation,
+                                                           NumEmbeddings>;
             float avg_time = 0;
             avg_time += launch_and_time_kernel(stream_config,
                                                kernel_main,
@@ -154,16 +140,13 @@ struct DeviceSparseEmbedding3ForwardLayernorm : public BaseOperator
                                                dim3(BlockSize),
                                                0,
                                                arg.p_out_,
-                                               arg.p_emb_a_,
-                                               arg.p_emb_b_,
-                                               arg.p_emb_c_,
-                                               arg.p_index_a_,
-                                               arg.p_index_b_,
-                                               arg.p_index_c_,
+                                               arg.p_embs_,
+                                               arg.p_indexs_,
                                                arg.p_gamma_,
                                                arg.p_beta_,
                                                out_desc,
-                                               arg.epsilon_);
+                                               arg.epsilon_,
+                                               arg.emb_elementwise_op_);
 
             return (avg_time);
         }
@@ -177,7 +160,7 @@ struct DeviceSparseEmbedding3ForwardLayernorm : public BaseOperator
 
     static bool IsSupportedArgument(const Argument* p_arg)
     {
-        return (RowPerBlock == p_arg->EmbeddingDim_) && (p_arg->NumRows_ % DimPerBlock == 0);
+        return (RowPerBlock == p_arg->EmbeddingDim_);
     }
 
     bool IsSupportedArgument(const BaseArgument* p_arg) override
@@ -195,7 +178,7 @@ struct DeviceSparseEmbedding3ForwardLayernorm : public BaseOperator
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DeviceSparseEmbedding3ForwardLayernorm_"<< BlockSize << "_" <<
+        str << "DeviceSparseEmbeddingsForwardLayernorm_"<< BlockSize << "_" <<
             DimClusterSize << "x" << RowClusterSize << "_" <<
             DimPerBlock << "x" << RowPerBlock << "_" <<
             DimThreadSize << "x" << RowVectorSize;
diff --git a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
similarity index 99%
rename from include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
index 8eab1cdee565b4334ede5ca9bb7544ea37ff498d..f849ac799d4611757e7a7c4d2949f602f5eb0e44 100644
--- a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -56,7 +56,8 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t num_blocks_per_batch =
@@ -938,7 +939,9 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/masking_specialization.hpp b/include/ck/tensor_operation/gpu/device/masking_specialization.hpp
index ea0f5897a757c408047b20e0cd4285e13f67c4f5..d6d6f74abdb5a63ef8a841d033c9f47ec59bd07e 100644
--- a/include/ck/tensor_operation/gpu/device/masking_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/masking_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/matrix_padder.hpp b/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
index 70e61bc7728f9638e5bef874ab0d5caad2db5028..c66d2fc516babe3aa5a16de053d9ada89eabf8ef 100644
--- a/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
+++ b/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp b/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp
index d35318357a9bcc6fb087c92e27dcf435ff60796b..5351d4ef2439307ad5623f4172217937812ab29a 100644
--- a/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp
+++ b/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/tensor_layout.hpp b/include/ck/tensor_operation/gpu/device/tensor_layout.hpp
index b44427411f9a28511fd16f6b47ee37fa84a0e2d3..b2d141fd61aa4550ef8b572e548688120634b8b9 100644
--- a/include/ck/tensor_operation/gpu/device/tensor_layout.hpp
+++ b/include/ck/tensor_operation/gpu/device/tensor_layout.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp b/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp
index 0ec0df2c9bbbb2aac987d681a36d5b00a8349c97..713fc93ebb4eec9ce0301affd667656bf0b6c999 100644
--- a/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/welford_helper.hpp b/include/ck/tensor_operation/gpu/device/welford_helper.hpp
index 6c909b767d4602914660d045b9aaf5a2307ea57b..d7772d8764ff3cc66572775f750997fae944800e 100644
--- a/include/ck/tensor_operation/gpu/device/welford_helper.hpp
+++ b/include/ck/tensor_operation/gpu/device/welford_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
index a4053b1f36221053bb2ef558dea469c64ce36e81..1dd96809d5f79ed4921fa6807625727279107df7 100644
--- a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
@@ -1,9 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck/utility/data_type.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -49,6 +50,14 @@ struct Add
         y = x0 + x1;
     };
 
+    template <>
+    __host__ __device__ constexpr void
+    operator()<float>(float& y, const float& x0, const bhalf_t& x1) const
+    {
+        const float x1_tmp = ck::type_convert<float>(x1);
+        y                  = x0 + x1_tmp;
+    }
+
     template <>
     __host__ __device__ constexpr void
     operator()<bhalf_t>(bhalf_t& y, const bhalf_t& x0, const bhalf_t& x1) const
@@ -67,6 +76,30 @@ struct Add
     };
 };
 
+struct ScaleAdd
+{
+    __host__ __device__ ScaleAdd(float scale) : scale_(scale) {}
+
+    template <typename Y, typename X0, typename X1>
+    __host__ __device__ constexpr void operator()(Y& y, const X0& x0, const X1& x1) const;
+
+    template <>
+    __host__ __device__ void
+    operator()<float, float, half_t>(float& y, const float& x0, const half_t& x1) const
+    {
+        y = scale_ * x0 + ck::type_convert<float>(x1);
+    };
+
+    template <>
+    __host__ __device__ void
+    operator()<float, float, bhalf_t>(float& y, const float& x0, const bhalf_t& x1) const
+    {
+        y = scale_ * x0 + ck::type_convert<float>(x1);
+    };
+
+    float scale_;
+};
+
 struct Subtract
 {
     template <typename T>
@@ -118,6 +151,13 @@ struct Bilinear
     template <typename Y, typename X0, typename X1>
     __host__ __device__ constexpr void operator()(Y&, const X0&, const X1&) const;
 
+    template <>
+    __host__ __device__ constexpr void
+    operator()<double, double, double>(double& y, const double& x0, const double& x1) const
+    {
+        y = alpha_ * x0 + beta_ * x1;
+    };
+
     template <>
     __host__ __device__ constexpr void
     operator()<float, float, float>(float& y, const float& x0, const float& x1) const
@@ -241,43 +281,42 @@ struct AddHardswish
     };
 };
 
-// C = A * B
 // E = FastGelu(C + D)
 struct AddFastGelu
 {
-    // Fast GeLU
-    // https://paperswithcode.com/method/gelu
-    // y = 0.5*x*(1+tanh(sqrt(2/pi)*(x+0.044715*x^3)))
-    __host__ __device__ static constexpr float GetFastGeLU(float x)
-    {
-        const float u   = 2.f * x * (0.035677f * x * x + 0.797885f);
-        const float emu = exp(-u);
-        const float cdf = 0.5f + 0.5f * (2.f / (1.f + emu) - 1.f);
-        return x * cdf;
-    }
-
-    template <typename T>
-    static inline constexpr bool is_valid_param_type_v =
-        std::is_same_v<T, float> || std::is_same_v<T, half_t> || std::is_same_v<T, bhalf_t> ||
-        std::is_same_v<T, int32_t> || std::is_same_v<T, int8_t>;
-
     template <typename E, typename C, typename D>
-    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void
+    operator()<float, float, float>(float& e, const float& c, const float& d) const
     {
-        static_assert(is_valid_param_type_v<E> && is_valid_param_type_v<C> &&
-                      is_valid_param_type_v<D>);
+        const float x = c + d;
 
-        const float y = GetFastGeLU(type_convert<float>(c) + type_convert<float>(d));
+        FastGelu{}.template operator()<float, float>(e, x);
+    }
 
-        e = type_convert<E>(y);
+    template <>
+    __host__ __device__ constexpr void
+    operator()<half_t, half_t, half_t>(half_t& e, const half_t& c, const half_t& d) const
+    {
+        const half_t x = c + d;
+
+        ck::tensor_operation::element_wise::FastGelu{}.template operator()<half_t, half_t>(e, x);
     }
 
-    template <typename D>
-    __host__ __device__ constexpr void operator()(float& e, const float& c, const D& d) const
+    template <>
+    __host__ __device__ constexpr void
+    operator()<half_t, float, half_t>(half_t& e, const float& c, const half_t& d) const
     {
-        static_assert(is_valid_param_type_v<D>);
+        const float x0_f = c + d;
+
+        float x1_f = 0;
+
+        ck::tensor_operation::element_wise::FastGelu{}.template operator()<float, float>(x1_f,
+                                                                                         x0_f);
 
-        e = GetFastGeLU(c + type_convert<float>(d));
+        e = type_convert<half_t>(x1_f);
     }
 };
 
diff --git a/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
index b66107a525569db2c8c706d7826221a686abf39e..3fdb391a0215e83328905b77148b675defc27af7 100644
--- a/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -16,7 +16,7 @@ namespace element_wise {
 // Need to ensure compiler will fail if there is no matching candidate, instead of compiler
 // siliently do implicit type conversion
 //
-// Method 1:
+// Example:
 //
 // struct ExampleElementwiseOp
 // {
@@ -30,19 +30,6 @@ namespace element_wise {
 //     {
 //     }
 // };
-//
-// Method 2:
-//
-// template <typename Y, typename X>
-// struct ExampleElementwiseOp;
-//
-// template <>
-// struct ExampleElementwiseOp<float, ck::bhalf_t>
-// {
-//     __host__ __device__ void operator()(float& y, ck::bhalf_t& x) const
-//     {
-//     }
-// };
 
 struct AddReluAdd
 {
@@ -173,40 +160,109 @@ struct AddAdd
 };
 
 // C = A * B
+// E = (C + D0) x D1
+struct AddMultiply
+{
+    template <typename E, typename C, typename D0, typename D1>
+    __host__ __device__ void operator()(E& e, const C& c, const D0& d0, const D1& d1) const;
+
+    template <>
+    __host__ __device__ void operator()<half_t, half_t, half_t, half_t>(half_t& e,
+                                                                        const half_t& c,
+                                                                        const half_t& d0,
+                                                                        const half_t& d1) const
+    {
+        const half_t y = (c + d0) * d1;
+        e              = y;
+    }
+    template <>
+    __host__ __device__ void operator()<half_t, float, half_t, half_t>(half_t& e,
+                                                                       const float& c,
+                                                                       const half_t& d0,
+                                                                       const half_t& d1) const
+    {
+        const half_t y = (type_convert<half_t>(c) + d0) * d1;
+        e              = y;
+    }
+    template <>
+    __host__ __device__ void operator()<float, float, half_t, half_t>(float& e,
+                                                                      const float& c,
+                                                                      const half_t& d0,
+                                                                      const half_t& d1) const
+    {
+        const float y = (c + d0) * d1;
+        e             = y;
+    }
+};
+
 // E = FastGelu(C + D0 + D1)
 struct AddAddFastGelu
 {
-    // Fast GeLU
-    // https://paperswithcode.com/method/gelu
-    // y = 0.5*x*(1+tanh(sqrt(2/pi)*(x+0.044715*x^3)))
-    __host__ __device__ static constexpr float GetFastGeLU(float x)
+    template <typename E, typename C, typename D0, typename D1>
+    __host__ __device__ constexpr void
+    operator()(E& e, const C& c, const D0& d0, const D1& d1) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<float, float, float, float>(float& e,
+                                                                              const float& c,
+                                                                              const float& d0,
+                                                                              const float& d1) const
     {
-        const float u   = 2.f * x * (0.035677f * x * x + 0.797885f);
-        const float emu = exp(-u);
-        const float cdf = 0.5f + 0.5f * (2.f / (1.f + emu) - 1.f);
-        return x * cdf;
+        const float x = c + d0 + d1;
+
+        FastGelu{}.template operator()<float, float>(e, x);
     }
 
-    template <typename T>
-    static inline constexpr bool is_valid_param_type_v =
-        std::is_same_v<T, float> || std::is_same_v<T, half_t> || std::is_same_v<T, bhalf_t> ||
-        std::is_same_v<T, int32_t> || std::is_same_v<T, int8_t>
-#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
-        || std::is_same_v<T, ck::int4_t>
-#endif
-        ;
+    template <>
+    __host__ __device__ constexpr void operator()<half_t, half_t, half_t, half_t>(
+        half_t& e, const half_t& c, const half_t& d0, const half_t& d1) const
+    {
+        const half_t x = c + d0 + d1;
 
-    template <typename E, typename C, typename D0, typename D1>
-    __host__ __device__ constexpr void
-    operator()(E& e, const C& c, const D0& d0, const D1& d1) const
+        ck::tensor_operation::element_wise::FastGelu{}.template operator()<half_t, half_t>(e, x);
+    }
+
+    template <>
+    __host__ __device__ constexpr void operator()<half_t, float, half_t, half_t>(
+        half_t& e, const float& c, const half_t& d0, const half_t& d1) const
+    {
+        const float x0_f = c + d0 + d1;
+
+        float x1_f = 0;
+
+        ck::tensor_operation::element_wise::FastGelu{}.template operator()<float, float>(x1_f,
+                                                                                         x0_f);
+
+        e = type_convert<half_t>(x1_f);
+    }
+
+    template <>
+    __host__ __device__ constexpr void operator()<bhalf_t, float, bhalf_t, bhalf_t>(
+        bhalf_t& e, const float& c, const bhalf_t& d0, const bhalf_t& d1) const
     {
-        static_assert(is_valid_param_type_v<E> && is_valid_param_type_v<C> &&
-                      is_valid_param_type_v<D0> && is_valid_param_type_v<D1>);
+        const float x0_f = c + type_convert<float>(d0) + type_convert<float>(d1);
 
-        const float y =
-            GetFastGeLU(type_convert<float>(c) + type_convert<float>(d0) + type_convert<float>(d1));
+        float x1_f = 0;
 
-        e = type_convert<E>(y);
+        ck::tensor_operation::element_wise::FastGelu{}.template operator()<float, float>(x1_f,
+                                                                                         x0_f);
+
+        e = type_convert<bhalf_t>(x1_f);
+    }
+
+    template <>
+    __host__ __device__ constexpr void operator()<int8_t, int32_t, int8_t, int8_t>(
+        int8_t& e, const int32_t& c, const int8_t& d0, const int8_t& d1) const
+    {
+        const float x0_f =
+            type_convert<float>(c) + type_convert<float>(d0) + type_convert<float>(d1);
+
+        float x1_f = 0;
+
+        ck::tensor_operation::element_wise::FastGelu{}.template operator()<float, float>(x1_f,
+                                                                                         x0_f);
+
+        e = type_convert<int8_t>(x1_f);
     }
 };
 
@@ -278,6 +334,40 @@ struct Normalize
     double epsilon_;
 };
 
+// used by BatchNorm inference
+// y = gamma * (x-mean) / sqrt(epsilon+variance) + beta
+// The data type of mean and variance is used as AccDataType
+struct NormalizeInInfer
+{
+    NormalizeInInfer(double epsilon = 1e-4) : epsilon_(epsilon) {}
+
+    template <typename T1, typename T2, typename T3, typename T4>
+    __host__ __device__ constexpr void operator()(T1& y,
+                                                  const T1& x,
+                                                  const T2& mean,
+                                                  const T2& variance,
+                                                  const T3& gamma,
+                                                  const T4& beta) const
+    {
+        static_assert(std::is_same<T2, float>::value || std::is_same<T2, double>::value,
+                      "Data type is not supported by this operation!");
+
+        using ck::type_convert;
+        using ck::math::sqrt;
+
+        T2 tmp_x, tmp_y;
+
+        tmp_x = type_convert<T2>(x);
+
+        tmp_y = ((tmp_x - mean) / sqrt(variance + type_convert<T2>(epsilon_))) *
+                    type_convert<T2>(gamma) +
+                type_convert<T2>(beta);
+        y = type_convert<T1>(tmp_y);
+    };
+
+    double epsilon_;
+};
+
 template <typename Y, typename X>
 struct UnaryTypeConvert;
 
diff --git a/include/ck/tensor_operation/gpu/element/quantization_operation.hpp b/include/ck/tensor_operation/gpu/element/quantization_operation.hpp
index 3f2c2f877314d07f90438bc59debf30c8aa84b8f..fefa6c793f7e91e3744fe028469fd4359e4e6bbc 100644
--- a/include/ck/tensor_operation/gpu/element/quantization_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/quantization_operation.hpp
@@ -1,15 +1,36 @@
 #pragma once
 
 #include "ck/utility/data_type.hpp"
+// #include "ck/utility/get_id.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace element_wise {
 
+// Y = Sy * Qy
+// W = Sw * Qw
+// X = Sx * Qx
+// B = Sb * Qb = Sw * Sx * Qb
+// Where X, W, Y are float32, Qx, Qw, Qy are int8
+// Sx, Sw, Sy are scale of x, w, y (float32), which is calculated from quantization range
+// Qb is int32, scale of B is Sw * Sx for convenient
+
+// Y = W @ X, where @ is convolution or matrix multiplication
+// Sy * Qy = Sw * Qw @ Sx * Qx
+// Qy = [(Sw*Sx)/Sy] * Qw @ Qx
+
 // For Activation function which is piecewise linear function, such as relu, leaky relu ...etc
+// Activation(Sy * Qy) = Sy * Activation(Qy)
 template <typename Activation>
 struct Activation_Mul_Clamp
 {
+    // Convolution + Activation (piecewise linear function)
+    // If an activation is piecewise linear function, then Activation(Sy * Qy) = Sy * Activation(Qy)
+    // Z = Activation(Y) = Activation(W @ X)
+    // Sz * Qz = Activation(Sy * Qy)
+    // Qz = Sy / Sz * Activation(Qy) = (Sw * Sx / Sz) * Activation(Qw @ Qx)
+
+    // requantScale_ = Sw * Sx / Sz
     Activation_Mul_Clamp(float requantScale, Activation activationOp)
         : requantScale_(requantScale), activationOp_(activationOp)
     {
@@ -17,26 +38,66 @@ struct Activation_Mul_Clamp
 
     __host__ __device__ constexpr void operator()(int8_t& y, const int32_t& x) const
     {
-        float x_fp32 = ck::type_convert<float>(x);
-        activationOp_(x_fp32, x_fp32);
-        float y_fp32 = math::clamp(requantScale_ * x_fp32, -128.f, 127.f);
-        y            = ck::type_convert<int8_t>(y_fp32);
+        float y_fp32 = ck::type_convert<float>(x);
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(requantScale_ * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int8_t>(y_fp32);
+    }
+
+    __device__ constexpr void operator()(int32_t& y, const int32_t& x) const
+    {
+        // CAUSION - We might type_convert to int8 in threadwise copy
+        // eg. GridwiseGemmDlMultipleD_km_kn_mn
+        float y_fp32 = ck::type_convert<float>(x);
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(requantScale_ * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int32_t>(y_fp32);
     }
 
-    __host__ __device__ constexpr void operator()(float& y, const int32_t& x) const
+    __host__ constexpr void operator()(float& y, const float& x) const
     {
-        // We might type_convert to int8 after lambda in someplace
-        float x_fp32 = ck::type_convert<float>(x);
-        activationOp_(x_fp32, x_fp32);
-        y = math::clamp(requantScale_ * x_fp32, -128.f, 127.f);
+        // CAUSION - We might float in & float out in reference code
+        activationOp_(y, x);
+        y = math::clamp(requantScale_ * y, -128.f, 127.f);
     }
 
     float requantScale_;
     Activation activationOp_;
 };
 
+// For Activation function which is non piecewise linear function, such as TanH, Sigmoid ...etc
+// If an activation is not piecewise linear function
+// then Activation(Sy * Qy) != Sy * Activation(Qy)
+template <typename Activation>
+struct Mul_Activation_Mul_Clamp
+{
+    // Convolution + Activation (non piecewise linear function)
+    // Z = Activation(Y) = Activation(W @ X)
+    // Sz * Qz = Activation(Sy * Qy)
+    // Qz = S1 * Activation[Sacc * (Qw @ Qx)]
+    // Where S1 = 1 / Sz, Sacc = Sw * Sx
+    Mul_Activation_Mul_Clamp(float scale_z_inv, float scaleAcc, Activation activationOp)
+        : scale_z_inv_(scale_z_inv), scaleAcc_(scaleAcc), activationOp_(activationOp)
+    {
+    }
+
+    __host__ __device__ constexpr void operator()(int8_t& y, const int32_t& x) const
+    {
+        float y_fp32 = ck::type_convert<float>(x);
+        y_fp32       = scaleAcc_ * y_fp32;
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(scale_z_inv_ * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int8_t>(y_fp32);
+    }
+
+    float scale_z_inv_;
+    float scaleAcc_;
+    Activation activationOp_;
+};
+
 // Conv Perchannel quantization + Activation function which is piecewise linear function, such as
 // relu, leaky relu ...etc
+// Activation(Sy * Qy) = Sy * Activation(Qy)
 template <typename Activation>
 struct Activation_Mul2_Clamp
 {
@@ -51,13 +112,35 @@ struct Activation_Mul2_Clamp
         y      = ck::type_convert<int8_t>(y_fp32);
     }
 
+    __device__ constexpr void
+    operator()(int32_t& y, const int32_t& x, const float& requantScale) const
+    {
+        // CAUSION - We might type_convert to int8 in threadwise copy
+        // eg. GridwiseGemmDlMultipleD_km_kn_mn
+        float y_fp32 = ck::type_convert<float>(x);
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(requantScale * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int32_t>(y_fp32);
+    }
+
     Activation activationOp_;
 };
 
 // For Activation function which is piecewise linear function, such as relu, leaky relu ...etc
+// Activation(Sy * Qy) = Sy * Activation(Qy)
 template <typename Activation>
 struct Add_Activation_Mul_Clamp
 {
+    // Convolution + bias
+    // Let Bias = B = Sw * Sx * Qb
+    // Where Qb is int32
+    // Y = W @ X + B
+    // Sy * Qy = Sw * Qw @ Sx * Qx + Sw * Sx * Qb
+    // Qy = [(Sw*Sx)/Sy] * (Qw @ Qx + Qb)
+
+    // For activation, Z = Activaiton(Y)
+    // Sz * Qz = Activation(Sy * Qy)
+    // Qz = Sy / Sz * Activation(Qy) = [(Sw*Sx)/Sz] * Activation(Qw @ Qx + Qb)
     Add_Activation_Mul_Clamp(float requantScale, Activation activationOp)
         : requantScale_(requantScale), activationOp_(activationOp)
     {
@@ -72,6 +155,17 @@ struct Add_Activation_Mul_Clamp
         y      = ck::type_convert<int8_t>(y_fp32);
     }
 
+    __host__ __device__ constexpr void
+    operator()(int32_t& y, const int32_t& x, const int32_t& bias) const
+    {
+        // CAUSION - We might type_convert to int8 in threadwise copy
+        // eg. GridwiseGemmDlMultipleD_km_kn_mn
+        float y_fp32 = ck::type_convert<float>(x + bias);
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(requantScale_ * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int32_t>(y_fp32);
+    }
+
     float requantScale_;
     Activation activationOp_;
 };
@@ -92,15 +186,33 @@ struct Add_Activation_Mul2_Clamp
         y      = ck::type_convert<int8_t>(y_fp32);
     }
 
+    __host__ __device__ constexpr void
+    operator()(int32_t& y, const int32_t& x, const int32_t& bias, const float& requantScale) const
+    {
+        // CAUSION - We might type_convert to int8 in threadwise copy
+        // eg. GridwiseGemmDlMultipleD_km_kn_mn
+        float y_fp32 = ck::type_convert<float>(x + bias);
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(requantScale * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int32_t>(y_fp32);
+    }
+
     Activation activationOp_;
 };
 
 // For Activation function which is non piecewise linear function, such as TanH, Sigmoid ...etc
+// If an activation is not piecewise linear function
+// then Activation(Sy * Qy) != Sy * Activation(Qy)
 template <typename Activation>
 struct Add_Mul_Activation_Mul_Clamp
 {
-    Add_Mul_Activation_Mul_Clamp(float requantScale1, float requantScale2, Activation activationOp)
-        : requantScale1_(requantScale1), requantScale2_(requantScale2), activationOp_(activationOp)
+    // Convolution + Activation (non piecewise linear function)
+    // Z = Activation(Y) = Activation(W @ X + B)
+    // Sz * Qz = Activation(Sy * Qy)
+    // Qz = S1 * Activation[Sacc * (Qw @ Qx + Qb)]
+    // Where S1 = 1 / Sz, Sacc = Sw * Sx
+    Add_Mul_Activation_Mul_Clamp(float scale_z_inv, float scaleAcc, Activation activationOp)
+        : scale_z_inv_(scale_z_inv), scaleAcc_(scaleAcc), activationOp_(activationOp)
     {
     }
 
@@ -108,14 +220,64 @@ struct Add_Mul_Activation_Mul_Clamp
     operator()(int8_t& y, const int32_t& x, const int32_t& bias) const
     {
         float y_fp32 = ck::type_convert<float>(x + bias);
-        y_fp32       = requantScale1_ * y_fp32;
+        y_fp32       = scaleAcc_ * y_fp32;
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(scale_z_inv_ * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int8_t>(y_fp32);
+    }
+
+    __host__ __device__ constexpr void
+    operator()(int32_t& y, const int32_t& x, const int32_t& bias) const
+    {
+        // CAUSION - We might type_convert to int8 in threadwise copy
+        // eg. GridwiseGemmDlMultipleD_km_kn_mn
+        float y_fp32 = ck::type_convert<float>(x + bias);
+        y_fp32       = scaleAcc_ * y_fp32;
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(scale_z_inv_ * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int32_t>(y_fp32);
+    }
+
+    float scale_z_inv_;
+    float scaleAcc_;
+    Activation activationOp_;
+};
+
+// Conv Perchannel quantization + Activation function which is non piecewise linear function,
+// such as TanH, Sigmoid ...etc
+// If an activation is not piecewise linear function
+// then Activation(Sy *Qy) != Sy * Activation(Qy)
+template <typename Activation>
+struct Add_Mul2_Activation_Mul_Clamp
+{
+    Add_Mul2_Activation_Mul_Clamp(float scale_z_inv, Activation activationOp)
+        : scale_z_inv_(scale_z_inv), activationOp_(activationOp)
+    {
+    }
+
+    __host__ __device__ constexpr void
+    operator()(int8_t& y, const int32_t& x, const int32_t& bias, const float& scaleAcc) const
+    {
+        float y_fp32 = ck::type_convert<float>(x + bias);
+        y_fp32       = scaleAcc * y_fp32;
         activationOp_(y_fp32, y_fp32);
-        y_fp32 = math::clamp(requantScale2_ * y_fp32, -128.f, 127.f);
+        y_fp32 = math::clamp(scale_z_inv_ * y_fp32, -128.f, 127.f);
         y      = ck::type_convert<int8_t>(y_fp32);
     }
 
-    float requantScale1_;
-    float requantScale2_;
+    __host__ __device__ constexpr void
+    operator()(int32_t& y, const int32_t& x, const int32_t& bias, const float& scaleAcc) const
+    {
+        // CAUSION - We might type_convert to int8 in threadwise copy
+        // eg. GridwiseGemmDlMultipleD_km_kn_mn
+        float y_fp32 = ck::type_convert<float>(x + bias);
+        y_fp32       = scaleAcc * y_fp32;
+        activationOp_(y_fp32, y_fp32);
+        y_fp32 = math::clamp(scale_z_inv_ * y_fp32, -128.f, 127.f);
+        y      = ck::type_convert<int32_t>(y_fp32);
+    }
+
+    float scale_z_inv_;
     Activation activationOp_;
 };
 
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 7f1cc22660b9df9390bee07fa8f171e812790dd3..8054e68d5b1d9aaea3457985d33bf62d3b4d193a 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -1,16 +1,21 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/math.hpp"
 #include "ck/utility/math_v2.hpp"
+#include "ck/utility/type_convert.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace element_wise {
 
+#if CK_WORKAROUND_SWDEV_383542
+extern "C" __device__ float __ocml_native_recip_f32(float);
+#endif
+
 struct PassThrough
 {
     template <typename Y, typename X>
@@ -52,6 +57,12 @@ struct PassThrough
         y = type_convert<bhalf_t>(x);
     }
 
+    template <>
+    __host__ __device__ void operator()<bhalf_t, half_t>(bhalf_t& y, const half_t& x) const
+    {
+        y = type_convert<bhalf_t>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<int8_t, int8_t>(int8_t& y, const int8_t& x) const
     {
@@ -71,6 +82,36 @@ struct PassThrough
         y = x;
     }
 #endif
+
+    template <>
+    __host__ __device__ void operator()<f8_t, f8_t>(f8_t& y, const f8_t& x) const
+    {
+        y = x;
+    }
+
+    template <>
+    __host__ __device__ void operator()<float, f8_t>(float& y, const f8_t& x) const
+    {
+        y = type_convert<float>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<f8_t, float>(f8_t& y, const float& x) const
+    {
+        y = type_convert<f8_t>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<half_t, f8_t>(half_t& y, const f8_t& x) const
+    {
+        y = type_convert<half_t>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<f8_t, half_t>(f8_t& y, const half_t& x) const
+    {
+        y = type_convert<f8_t>(x);
+    }
 };
 
 struct UnaryConvert
@@ -82,6 +123,40 @@ struct UnaryConvert
     }
 };
 
+struct ConvertBF16RTN
+{
+    // convert to bf16 using round to nearest (rtn)
+    template <typename Y, typename X>
+    __host__ __device__ void operator()(Y& y, const X& x) const
+    {
+        // check Y datatype
+        static_assert(is_same<Y, bhalf_t>::value, "Data type is not supported by this operation!");
+
+        // check X datatype
+        static_assert(is_same<X, float>::value || is_same<X, half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        y = bf16_convert_rtn<Y>(x);
+    }
+};
+
+struct ConvertF8SR
+{
+    // convert to fp8 using stochastic rounding (SR)
+    template <typename Y, typename X>
+    __host__ __device__ void operator()(Y& y, const X& x) const
+    {
+        // check Y datatype
+        static_assert(is_same<Y, f8_t>::value, "Data type is not supported by this operation!");
+
+        // check X datatype
+        static_assert(is_same<X, float>::value || is_same<X, half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        y = f8_convert_sr<Y>(x);
+    }
+};
+
 struct Scale
 {
     __host__ __device__ Scale(float scale) : scale_(scale) {}
@@ -95,6 +170,12 @@ struct Scale
         y = scale_ * x;
     };
 
+    template <>
+    __host__ __device__ void operator()<double, double>(double& y, const double& x) const
+    {
+        y = scale_ * x;
+    };
+
     __host__ __device__ auto Value() const { return scale_; }
 
     float scale_;
@@ -196,36 +277,83 @@ struct Relu
     }
 };
 
-// Y = FastGelu(X)
+// Fast GeLU
+// https://paperswithcode.com/method/gelu
+// y = 0.5*x*(1+tanh(sqrt(2/pi)*(x+0.044715*x^3)))
+// host code use higher accuracy "exp" and "div"
+// gpu code use lower accuracy "__expf" and "rcp" function
 struct FastGelu
 {
-    // Fast GeLU
-    // https://paperswithcode.com/method/gelu
-    // y = 0.5*x*(1+tanh(sqrt(2/pi)*(x+0.044715*x^3)))
-    __host__ __device__ static constexpr float GetFastGeLU(float x)
+    template <typename Y, typename X>
+    __host__ void operator()(Y& y, const X& x) const;
+
+    template <typename Y, typename X>
+    __device__ void operator()(Y& y, const X& x) const;
+
+    template <>
+    __host__ void operator()<float, float>(float& y, const float& x) const
     {
         const float u   = 2.f * x * (0.035677f * x * x + 0.797885f);
         const float emu = exp(-u);
         const float cdf = 0.5f + 0.5f * (2.f / (1.f + emu) - 1.f);
-        return x * cdf;
+
+        y = x * cdf;
     }
 
-    template <typename T>
-    static inline constexpr bool is_valid_param_type_v =
-        std::is_same_v<T, float> || std::is_same_v<T, half_t> || std::is_same_v<T, bhalf_t> ||
-        std::is_same_v<T, int32_t> || std::is_same_v<T, int8_t>
-#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
-        || std::is_same_v<T, ck::int4_t>
+    // device code, use lower precision "__expf" and "rcp"
+    template <>
+    __device__ void operator()<float, float>(float& y, const float& x) const
+    {
+        const float u   = 2.f * x * (0.035677f * x * x + 0.797885f);
+        const float emu = __expf(-u);
+
+#if !CK_WORKAROUND_SWDEV_383542
+        const float cdf = 0.5f + 0.5f * (2.f * __frcp_rn(1.f + emu) - 1.f);
+#else
+        const float cdf = 0.5f + 0.5f * (2.f * __ocml_native_recip_f32(1.f + emu) - 1.f);
 #endif
-        ;
 
-    template <typename Y, typename X>
-    __host__ __device__ void operator()(Y& y, const X& x) const
+        y = x * cdf;
+    }
+
+    template <>
+    __host__ void operator()<half_t, half_t>(half_t& y, const half_t& x) const
     {
-        static_assert(is_valid_param_type_v<Y> && is_valid_param_type_v<X>);
+        float y_f;
+
+        this->operator()<float, float>(y_f, type_convert<float>(x));
 
-        const float tmp_y = GetFastGeLU(type_convert<float>(x));
-        y                 = type_convert<Y>(tmp_y);
+        y = type_convert<half_t>(y_f);
+    }
+
+    template <>
+    __device__ void operator()<half_t, half_t>(half_t& y, const half_t& x) const
+    {
+        float y_f;
+
+        this->operator()<float, float>(y_f, type_convert<float>(x));
+
+        y = type_convert<half_t>(y_f);
+    }
+
+    template <>
+    __host__ void operator()<half_t, float>(half_t& y, const float& x) const
+    {
+        float y_f;
+
+        this->operator()<float, float>(y_f, x);
+
+        y = type_convert<half_t>(y_f);
+    }
+
+    template <>
+    __device__ void operator()<half_t, float>(half_t& y, const float& x) const
+    {
+        float y_f;
+
+        this->operator()<float, float>(y_f, x);
+
+        y = type_convert<half_t>(y_f);
     }
 };
 
@@ -261,8 +389,36 @@ struct Sigmoid
 
         y = 1 / (ck::type_convert<T>(1) + exp(-x));
     };
+};
 
-    int32_t divider_ = 1;
+struct TanH
+{
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
+                          is_same<T, ck::half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        y = ck::math::tanh(x);
+    };
+};
+
+struct Swish
+{
+    Swish(float beta = 1.0f) : beta_(beta) {}
+
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
+                          is_same<T, ck::half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        y = x / (ck::type_convert<T>(1) + ck::math::exp(-beta_ * x));
+    };
+
+    float beta_ = 1.0f;
 };
 
 } // namespace element_wise
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..f73fe282db98a0e88120a73b89a1a071fcbb2dea
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
@@ -0,0 +1,704 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/math_v2.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/utility/workgroup_synchronization.hpp"
+
+namespace ck {
+
+template <typename GridwiseMultiblockBatchNormForward_,
+          typename XDataType,
+          typename YDataType,
+          typename AccDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          typename YElementwiseOp,
+          typename XYGridDesc_M_K,
+          typename MeanVarCountGridDesc_M_G,
+          typename MeanVarCountGridDesc_M_K,
+          typename ScaleBiasGridDesc_M,
+          typename MeanVarGridDesc_M,
+          typename GetReduceCountPerThreadFunctor>
+__global__ void kernel_multiblock_batchnorm_forward(
+    const XYGridDesc_M_K x_grid_desc_m_k,
+    const XYGridDesc_M_K y_grid_desc_m_k,
+    const MeanVarCountGridDesc_M_G mean_var_count_grid_desc_m_g,
+    const MeanVarCountGridDesc_M_K mean_var_count_grid_desc_m_k,
+    const ScaleBiasGridDesc_M scale_grid_desc_m,
+    const ScaleBiasGridDesc_M bias_grid_desc_m,
+    const MeanVarGridDesc_M mean_var_grid_desc_m,
+    const GetReduceCountPerThreadFunctor get_reduce_count_per_thread,
+    index_t num_k_block_tile_iteration,
+    AccDataType epsilon,
+    const XDataType* const __restrict__ p_x,
+    MeanVarDataType* const __restrict__ p_welford_mean,
+    MeanVarDataType* const __restrict__ p_welford_variance,
+    int32_t* const __restrict__ p_welford_count,
+    int32_t* const __restrict__ p_control,
+    const ScaleDataType* const __restrict__ p_scale,
+    const BiasDataType* const __restrict__ p_bias,
+    const YElementwiseOp y_elementwise_op,
+    YDataType* const __restrict__ p_y,
+    bool updateMovingAverage,
+    AccDataType averageFactor,
+    MeanVarDataType* const __restrict__ resultRunningMean,
+    MeanVarDataType* const __restrict__ resultRunningVariance,
+    bool saveMeanInvVariance,
+    MeanVarDataType* const __restrict__ resultSaveMean,
+    MeanVarDataType* const __restrict__ resultSaveInvVariance)
+{
+    GridwiseMultiblockBatchNormForward_::Run(x_grid_desc_m_k,
+                                             y_grid_desc_m_k,
+                                             mean_var_count_grid_desc_m_g,
+                                             mean_var_count_grid_desc_m_k,
+                                             scale_grid_desc_m,
+                                             bias_grid_desc_m,
+                                             mean_var_grid_desc_m,
+                                             get_reduce_count_per_thread,
+                                             num_k_block_tile_iteration,
+                                             epsilon,
+                                             p_x,
+                                             p_welford_mean,
+                                             p_welford_variance,
+                                             p_welford_count,
+                                             p_control,
+                                             p_scale,
+                                             p_bias,
+                                             y_elementwise_op,
+                                             p_y,
+                                             updateMovingAverage,
+                                             averageFactor,
+                                             resultRunningMean,
+                                             resultRunningVariance,
+                                             saveMeanInvVariance,
+                                             resultSaveMean,
+                                             resultSaveInvVariance);
+};
+
+template <typename XDataType,
+          typename YDataType,
+          typename AccDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          typename YElementwiseOp,
+          typename XYGridDesc_M_K,
+          typename MeanVarCountGridDesc_M_G,
+          typename MeanVarCountGridDesc_M_K,
+          typename ScaleBiasGridDesc_M,
+          typename MeanVarGridDesc_M,
+          typename GetReduceCountPerThreadFunctor,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcYDstVectorDim,
+          index_t XSrcVectorSize,
+          index_t YDstVectorSize,
+          index_t ScaleSrcVectorSize,
+          index_t BiasSrcVectorSize,
+          index_t MeanVarSrcDstVectorSize>
+struct GridwiseMultiblockBatchNormForward
+{
+    static_assert((XSrcYDstVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcYDstVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert((XSrcYDstVectorDim == 0 && MThreadSliceSize % YDstVectorSize == 0) ||
+                      (XSrcYDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static constexpr bool reorder_thread_cluster = (XSrcYDstVectorDim == 0);
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using ThreadBufferDimAccessOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{})));
+    using ThreadReduceDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using ThreadReduceSrcDesc_M_1 = decltype(
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
+
+    using ThreadwiseWelford1 =
+        ThreadwiseWelford<AccDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
+
+    using ThreadwiseWelford2 =
+        ThreadwiseWelfordMerge<AccDataType, ThreadReduceSrcDesc_M_1, ThreadReduceDstDesc_M>;
+
+    using BlockwiseWelford1 = BlockwiseWelford<AccDataType,
+                                               BlockSize,
+                                               ThreadClusterLengths_M_K,
+                                               ThreadClusterArrangeOrder,
+                                               false>;
+
+    using BlockwiseWelford2 = BlockwiseWelford<AccDataType,
+                                               BlockSize,
+                                               ThreadClusterLengths_M_K,
+                                               ThreadClusterArrangeOrder,
+                                               true>;
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    __device__ static void Run(const XYGridDesc_M_K& x_grid_desc_m_k,
+                               const XYGridDesc_M_K& y_grid_desc_m_k,
+                               const MeanVarCountGridDesc_M_G& mean_var_count_grid_desc_m_g,
+                               const MeanVarCountGridDesc_M_K& mean_var_count_grid_desc_m_k,
+                               const ScaleBiasGridDesc_M& scale_grid_desc_m,
+                               const ScaleBiasGridDesc_M& bias_grid_desc_m,
+                               const MeanVarGridDesc_M& mean_var_grid_desc_m,
+                               const GetReduceCountPerThreadFunctor& get_reduce_count_per_thread,
+                               index_t num_k_block_tile_iteration,
+                               AccDataType epsilon,
+                               const XDataType* const __restrict__ p_x,
+                               MeanVarDataType* const __restrict__ p_welford_mean,
+                               MeanVarDataType* const __restrict__ p_welford_variance,
+                               int32_t* const __restrict__ p_welford_count,
+                               int32_t* const __restrict__ p_control,
+                               const ScaleDataType* const __restrict__ p_scale,
+                               const BiasDataType* const __restrict__ p_bias,
+                               const YElementwiseOp y_elementwise_op,
+                               YDataType* const __restrict__ p_y,
+                               bool updateMovingAverage,
+                               AccDataType averageFactor,
+                               MeanVarDataType* const __restrict__ resultRunningMean,
+                               MeanVarDataType* const __restrict__ resultRunningVariance,
+                               bool saveMeanInvVariance,
+                               MeanVarDataType* const __restrict__ resultSaveMean,
+                               MeanVarDataType* const __restrict__ resultSaveInvVariance)
+    {
+        using ck::math::sqrt;
+
+        const index_t blkgroup_size = mean_var_count_grid_desc_m_g.GetLength(I1);
+
+        const index_t thread_local_id = get_thread_local_1d_id();
+        const index_t block_global_id = get_block_1d_id();
+        const index_t blkgroup_id     = block_global_id / blkgroup_size;
+        const index_t block_local_id  = block_global_id % blkgroup_size;
+
+        if(block_local_id == 0)
+            gms_init(BlockSize / warpSize * blkgroup_size, &p_control[blkgroup_id * 2]);
+
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        using ThreadBufferLengths_M_K = Sequence<MThreadSliceSize, KThreadSliceSize>;
+        using ThreadBufferLengths_M   = Sequence<MThreadSliceSize>;
+        using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
+
+        constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+            make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{}));
+        constexpr auto thread_buffer_desc_m =
+            make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}));
+        constexpr auto thread_buffer_desc_m_1 = make_naive_tensor_descriptor_packed(
+            make_tuple(Number<MThreadSliceSize>{}, Number<1>{}));
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize * KThreadSliceSize, true>
+            x_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true> count_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+            tmp_mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+            tmp_var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true> tmp_count_thread_buf;
+
+        const index_t reduceSizePerBlock = K_BlockTileSize * num_k_block_tile_iteration;
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  AccDataType,
+                                                                  XYGridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  ThreadBufferDimAccessOrder,
+                                                                  XSrcYDstVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(blkgroup_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                             block_local_id * reduceSizePerBlock +
+                                 thread_k_cluster_id * KThreadSliceSize));
+
+        constexpr auto xy_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileSize);
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x, x_grid_desc_m_k.GetElementSpaceSize());
+
+        // Step 1: each workgroup does local welford reduction
+
+        auto threadwise_welford_1 = ThreadwiseWelford1();
+        threadwise_welford_1.max_count_ =
+            get_reduce_count_per_thread(block_local_id, thread_k_cluster_id);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I) = type_convert<AccDataType>(0.0f);
+            var_thread_buf(I)  = type_convert<AccDataType>(0.0f);
+        });
+
+        for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+        {
+            threadwise_x_load.Run(x_grid_desc_m_k,
+                                  x_global_val_buf,
+                                  thread_buffer_desc_m_k,
+                                  make_tuple(I0, I0),
+                                  x_thread_buf);
+
+            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, xy_copy_fwd_step_m_k);
+            threadwise_welford_1.Run(x_thread_buf, mean_thread_buf, var_thread_buf);
+        }
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            count_thread_buf(I) = threadwise_welford_1.cur_count_;
+            BlockwiseWelford1::Run(mean_thread_buf(I), var_thread_buf(I), count_thread_buf(I));
+        });
+
+        // Step 2: each workgroup writes its local welford result to workspace memory
+
+        auto mean_global_val_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global, AmdBufferCoherenceEnum::GLC>(
+                p_welford_mean, mean_var_count_grid_desc_m_g.GetElementSpaceSize());
+
+        auto var_global_val_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global, AmdBufferCoherenceEnum::GLC>(
+                p_welford_variance, mean_var_count_grid_desc_m_g.GetElementSpaceSize());
+
+        auto count_global_val_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global, AmdBufferCoherenceEnum::GLC>(
+                p_welford_count, mean_var_count_grid_desc_m_g.GetElementSpaceSize());
+
+        auto threadwise_mean_var_store_m_g =
+            ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                               MeanVarDataType,
+                                               decltype(thread_buffer_desc_m_1),
+                                               MeanVarCountGridDesc_M_G,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M_1,
+                                               Sequence<0, 1>,
+                                               0,
+                                               1,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                mean_var_count_grid_desc_m_g,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_local_id),
+                PassThroughOp{});
+
+        auto threadwise_count_store_m_g =
+            ThreadwiseTensorSliceTransfer_v1r3<int32_t,
+                                               int32_t,
+                                               decltype(thread_buffer_desc_m_1),
+                                               MeanVarCountGridDesc_M_G,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M_1,
+                                               Sequence<0, 1>,
+                                               0,
+                                               1,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                mean_var_count_grid_desc_m_g,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_local_id),
+                PassThroughOp{});
+
+        if(thread_k_cluster_id == 0)
+        {
+            threadwise_mean_var_store_m_g.Run(thread_buffer_desc_m_1,
+                                              make_tuple(I0, I0),
+                                              mean_thread_buf,
+                                              mean_var_count_grid_desc_m_g,
+                                              mean_global_val_buf);
+
+            threadwise_mean_var_store_m_g.Run(thread_buffer_desc_m_1,
+                                              make_tuple(I0, I0),
+                                              var_thread_buf,
+                                              mean_var_count_grid_desc_m_g,
+                                              var_global_val_buf);
+
+            threadwise_count_store_m_g.Run(thread_buffer_desc_m_1,
+                                           make_tuple(I0, I0),
+                                           count_thread_buf,
+                                           mean_var_count_grid_desc_m_g,
+                                           count_global_val_buf);
+        };
+
+        gms_barrier(&p_control[blkgroup_id * 2]);
+
+        if(block_local_id == 0)
+            gms_reset(&p_control[blkgroup_id * 2]);
+
+        // Step 3: each workgroup reads welford results from workspace memory and does final welford
+        // reduction
+
+        auto threadwise_mean_var_load_m_k =
+            ThreadwiseTensorSliceTransfer_v2<MeanVarDataType,
+                                             AccDataType,
+                                             MeanVarCountGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             0,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_count_grid_desc_m_k,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * 1));
+
+        auto threadwise_count_load_m_k =
+            ThreadwiseTensorSliceTransfer_v2<int32_t,
+                                             int32_t,
+                                             MeanVarCountGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             0,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_count_grid_desc_m_k,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * 1));
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I)  = type_convert<AccDataType>(0.0f);
+            var_thread_buf(I)   = type_convert<AccDataType>(0.0f);
+            count_thread_buf(I) = 0;
+        });
+
+        constexpr auto mean_var_count_read_fwd_step_m_k = make_multi_index(0, KThreadClusterSize);
+
+        int32_t reducedSize = 0;
+        while(reducedSize < blkgroup_size)
+        {
+            threadwise_mean_var_load_m_k.Run(mean_var_count_grid_desc_m_k,
+                                             mean_global_val_buf,
+                                             thread_buffer_desc_m_1,
+                                             make_tuple(I0, I0),
+                                             tmp_mean_thread_buf);
+
+            threadwise_mean_var_load_m_k.Run(mean_var_count_grid_desc_m_k,
+                                             var_global_val_buf,
+                                             thread_buffer_desc_m_1,
+                                             make_tuple(I0, I0),
+                                             tmp_var_thread_buf);
+
+            threadwise_count_load_m_k.Run(mean_var_count_grid_desc_m_k,
+                                          count_global_val_buf,
+                                          thread_buffer_desc_m_1,
+                                          make_tuple(I0, I0),
+                                          tmp_count_thread_buf);
+
+            ThreadwiseWelford2::Run(tmp_mean_thread_buf,
+                                    tmp_var_thread_buf,
+                                    tmp_count_thread_buf,
+                                    mean_thread_buf,
+                                    var_thread_buf,
+                                    count_thread_buf);
+
+            reducedSize += KThreadClusterSize;
+
+            threadwise_mean_var_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
+                                                            mean_var_count_read_fwd_step_m_k);
+            threadwise_count_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
+                                                         mean_var_count_read_fwd_step_m_k);
+        };
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            BlockwiseWelford2::Run(mean_thread_buf(I), var_thread_buf(I), count_thread_buf(I));
+        });
+
+        // Step 4: do normalization using the mean/variance
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> scale_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> bias_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize * KThreadSliceSize, true>
+            y_thread_buf;
+
+        auto threadwise_y_store =
+            ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                               YDataType,
+                                               decltype(thread_buffer_desc_m_k),
+                                               XYGridDesc_M_K,
+                                               YElementwiseOp,
+                                               ThreadBufferLengths_M_K,
+                                               ThreadBufferDimAccessOrder,
+                                               XSrcYDstVectorDim,
+                                               YDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                y_grid_desc_m_k,
+                make_multi_index(
+                    blkgroup_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                    block_local_id * reduceSizePerBlock + thread_k_cluster_id * KThreadSliceSize),
+                y_elementwise_op);
+
+        auto threadwise_scale_load =
+            ThreadwiseTensorSliceTransfer_v2<ScaleDataType,
+                                             AccDataType,
+                                             ScaleBiasGridDesc_M,
+                                             decltype(thread_buffer_desc_m),
+                                             ThreadBufferLengths_M,
+                                             Sequence<0>,
+                                             0,
+                                             ScaleSrcVectorSize,
+                                             1,
+                                             true>(
+                scale_grid_desc_m,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize));
+
+        auto threadwise_bias_load = ThreadwiseTensorSliceTransfer_v2<BiasDataType,
+                                                                     AccDataType,
+                                                                     ScaleBiasGridDesc_M,
+                                                                     decltype(thread_buffer_desc_m),
+                                                                     ThreadBufferLengths_M,
+                                                                     Sequence<0>,
+                                                                     0,
+                                                                     BiasSrcVectorSize,
+                                                                     1,
+                                                                     true>(
+            bias_grid_desc_m,
+            make_multi_index(blkgroup_id * M_BlockTileSize +
+                             thread_m_cluster_id * MThreadSliceSize));
+
+        const auto scale_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_scale, scale_grid_desc_m.GetElementSpaceSize());
+
+        const auto bias_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_bias, bias_grid_desc_m.GetElementSpaceSize());
+
+        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_y, y_grid_desc_m_k.GetElementSpaceSize());
+
+        threadwise_scale_load.Run(scale_grid_desc_m,
+                                  scale_global_val_buf,
+                                  thread_buffer_desc_m,
+                                  make_tuple(I0),
+                                  scale_thread_buf);
+
+        threadwise_bias_load.Run(bias_grid_desc_m,
+                                 bias_global_val_buf,
+                                 thread_buffer_desc_m,
+                                 make_tuple(I0),
+                                 bias_thread_buf);
+
+        threadwise_x_load.SetSrcSliceOrigin(
+            x_grid_desc_m_k,
+            make_multi_index(blkgroup_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                             block_local_id * reduceSizePerBlock +
+                                 thread_k_cluster_id * KThreadSliceSize));
+
+        for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+        {
+            threadwise_x_load.Run(x_grid_desc_m_k,
+                                  x_global_val_buf,
+                                  thread_buffer_desc_m_k,
+                                  make_tuple(I0, I0),
+                                  x_thread_buf);
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                AccDataType multiplier =
+                    scale_thread_buf[Number<iM>{}] / sqrt(var_thread_buf[iM] + epsilon);
+
+                AccDataType fused_mean_bias =
+                    bias_thread_buf[Number<iM>{}] - mean_thread_buf[iM] * multiplier;
+
+                static_for<0, KThreadSliceSize, 1>{}([&](auto iK) {
+                    constexpr auto offset =
+                        thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK));
+
+                    // normalize
+                    y_thread_buf(Number<offset>{}) =
+                        x_thread_buf[Number<offset>{}] * multiplier + fused_mean_bias;
+                });
+            });
+
+            threadwise_y_store.Run(thread_buffer_desc_m_k,
+                                   make_tuple(I0, I0),
+                                   y_thread_buf,
+                                   y_grid_desc_m_k,
+                                   y_global_val_buf);
+
+            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, xy_copy_fwd_step_m_k);
+            threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, xy_copy_fwd_step_m_k);
+        }
+
+        // Step 5: update the moving average of mean and variance (optional)
+
+        if(updateMovingAverage && block_local_id == 0 && thread_k_cluster_id == 0)
+        {
+            StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+                running_mean_thread_buf;
+            StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+                running_var_thread_buf;
+
+            auto running_mean_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultRunningMean, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            auto running_var_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultRunningVariance, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            auto threadwise_mean_var_load =
+                ThreadwiseTensorSliceTransfer_v2<MeanVarDataType,
+                                                 AccDataType,
+                                                 MeanVarGridDesc_M,
+                                                 decltype(thread_buffer_desc_m),
+                                                 ThreadBufferLengths_M,
+                                                 Sequence<0>,
+                                                 0,
+                                                 MeanVarSrcDstVectorSize,
+                                                 1,
+                                                 true>(
+                    mean_var_grid_desc_m,
+                    make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize));
+
+            threadwise_mean_var_load.Run(mean_var_grid_desc_m,
+                                         running_mean_global_buf,
+                                         thread_buffer_desc_m,
+                                         make_tuple(I0),
+                                         running_mean_thread_buf);
+
+            threadwise_mean_var_load.Run(mean_var_grid_desc_m,
+                                         running_var_global_buf,
+                                         thread_buffer_desc_m,
+                                         make_tuple(I0),
+                                         running_var_thread_buf);
+
+            AccDataType oneMinusAverageFactor = type_convert<AccDataType>(1.0) - averageFactor;
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                running_mean_thread_buf(I) = running_mean_thread_buf[I] * oneMinusAverageFactor +
+                                             mean_thread_buf[I] * averageFactor;
+                running_var_thread_buf(I) = running_var_thread_buf[I] * oneMinusAverageFactor +
+                                            var_thread_buf[I] * averageFactor;
+            });
+
+            auto threadwise_mean_var_store =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   MeanVarDataType,
+                                                   decltype(thread_buffer_desc_m),
+                                                   MeanVarGridDesc_M,
+                                                   PassThroughOp,
+                                                   ThreadBufferLengths_M,
+                                                   Sequence<0>,
+                                                   0,
+                                                   MeanVarSrcDstVectorSize,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>(
+                    mean_var_grid_desc_m,
+                    make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize),
+                    PassThroughOp{});
+
+            threadwise_mean_var_store.Run(thread_buffer_desc_m,
+                                          make_tuple(I0),
+                                          running_mean_thread_buf,
+                                          mean_var_grid_desc_m,
+                                          running_mean_global_buf);
+
+            threadwise_mean_var_store.Run(thread_buffer_desc_m,
+                                          make_tuple(I0),
+                                          running_var_thread_buf,
+                                          mean_var_grid_desc_m,
+                                          running_var_global_buf);
+        };
+
+        // Step 6: save mean and inv-variance (optional)
+
+        if(saveMeanInvVariance && block_local_id == 0 && thread_k_cluster_id == 0)
+        {
+            auto result_mean_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultSaveMean, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            auto result_inv_var_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultSaveInvVariance, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            // calculate inv-variance as 1/sqrt(epsilon+variance), stored in place of variance
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                var_thread_buf(I) =
+                    type_convert<AccDataType>(1.0f) / sqrt(epsilon + var_thread_buf[I]);
+            });
+
+            auto threadwise_mean_inv_var_store =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   MeanVarDataType,
+                                                   decltype(thread_buffer_desc_m),
+                                                   MeanVarGridDesc_M,
+                                                   PassThroughOp,
+                                                   ThreadBufferLengths_M,
+                                                   Sequence<0>,
+                                                   0,
+                                                   MeanVarSrcDstVectorSize,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>(
+                    mean_var_grid_desc_m,
+                    make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize),
+                    PassThroughOp{});
+
+            threadwise_mean_inv_var_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              mean_thread_buf,
+                                              mean_var_grid_desc_m,
+                                              result_mean_global_buf);
+
+            threadwise_mean_inv_var_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              var_thread_buf,
+                                              mean_var_grid_desc_m,
+                                              result_inv_var_global_buf);
+        };
+    }
+}; // namespace ck
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
index a72a4ee068f2bbcb1a915df0368c9b05caa298f9..e73e7e681721c4558c7b58b0bce962b6e156ec58 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
index 08cb0dd191303813a9283b19e38e2952fa0ccf27..a82a173500d21305cf5b643553d4ddf544a62d0d 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -161,7 +161,7 @@ struct GridwiseMultiblockWelfordFirstHalf
                                                PassThroughOp,
                                                ThreadBufferLengths_M_1,
                                                Sequence<0, 1>,
-                                               1,
+                                               0,
                                                1,
                                                InMemoryDataOperationEnum::Set,
                                                1,
@@ -180,7 +180,7 @@ struct GridwiseMultiblockWelfordFirstHalf
                                                PassThroughOp,
                                                ThreadBufferLengths_M_1,
                                                Sequence<0, 1>,
-                                               1,
+                                               0,
                                                1,
                                                InMemoryDataOperationEnum::Set,
                                                1,
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
similarity index 98%
rename from include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp
rename to include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
index 548d7fd40ac70fe7fde5e5973a177600b0941c5c..dd514567d243218784ef00363d5ecf9bc00bd1aa 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -33,7 +33,6 @@ __global__ void kernel_welford_second_half_batchnorm_forward_final(
     const MeanVarGridDesc_M mean_var_grid_desc_m,
     index_t blkgroup_size,
     index_t num_xy_k_block_tile_iteration,
-    index_t num_mean_var_count_k_block_tile_iteration,
     AccDataType epsilon,
     const MeanVarDataType* const __restrict__ p_in_welford_mean,
     const MeanVarDataType* const __restrict__ p_in_welford_variance,
@@ -59,7 +58,6 @@ __global__ void kernel_welford_second_half_batchnorm_forward_final(
                                                          mean_var_grid_desc_m,
                                                          blkgroup_size,
                                                          num_xy_k_block_tile_iteration,
-                                                         num_mean_var_count_k_block_tile_iteration,
                                                          epsilon,
                                                          p_in_welford_mean,
                                                          p_in_welford_variance,
@@ -152,7 +150,6 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                const MeanVarGridDesc_M& mean_var_grid_desc_m,
                                index_t blkgroup_size,
                                index_t num_xy_k_block_tile_iteration,
-                               index_t num_mean_var_count_k_block_tile_iteration,
                                AccDataType epsilon,
                                const MeanVarDataType* const __restrict__ p_in_welford_mean,
                                const MeanVarDataType* const __restrict__ p_in_welford_variance,
@@ -223,7 +220,7 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                              decltype(thread_buffer_desc_m_1),
                                              ThreadBufferLengths_M_1,
                                              Sequence<0, 1>,
-                                             1,
+                                             0,
                                              1,
                                              1,
                                              true>(
@@ -239,7 +236,7 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                              decltype(thread_buffer_desc_m_1),
                                              ThreadBufferLengths_M_1,
                                              Sequence<0, 1>,
-                                             1,
+                                             0,
                                              1,
                                              1,
                                              true>(
@@ -257,9 +254,6 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
         const auto welford_count_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_in_welford_count, mean_var_count_grid_desc_m_k.GetElementSpaceSize());
 
-        constexpr auto mean_var_count_thread_copy_step_m_k =
-            make_multi_index(0, KThreadClusterSize * 1);
-
         // Step 1: do final welford reduction to get mean and variance
 
         static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
@@ -268,8 +262,11 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
             welford_count_thread_buf(I) = 0;
         });
 
-        for(index_t reducedTiles = 0; reducedTiles < num_mean_var_count_k_block_tile_iteration;
-            ++reducedTiles)
+        constexpr auto mean_var_count_thread_copy_step_m_k =
+            make_multi_index(0, KThreadClusterSize);
+
+        int32_t reducedSize = 0;
+        while(reducedSize < blkgroup_size)
         {
             threadwise_mean_var_load_m_k.Run(mean_var_count_grid_desc_m_k,
                                              welford_mean_global_val_buf,
@@ -296,6 +293,8 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                    welford_var_thread_buf,
                                    welford_count_thread_buf);
 
+            reducedSize += KThreadClusterSize;
+
             threadwise_mean_var_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
                                                             mean_var_count_thread_copy_step_m_k);
             threadwise_count_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
index 42b7e172b239e69ae0ca840dd80d8bb203a5480f..6fe78edb34df290de485f0d015c3ccd86e82b65b 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index 460b684346b531db5381543372b01d6fe3e2f52e..5c8b9f419ae11d62a23f83320a8659122fb5d52d 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -109,30 +109,57 @@ struct BlockToCTileMap_M00_N0_M01
 
 // Rows of column-vectors
 // This C-tile map dynamically adjusts M01 when C-tile index is out of range
-template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N>
-struct BlockToCTileMap_M00_N0_M01Adapt
+template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N = void>
+struct BlockToCTileMap_M00_N0_M01Adapt;
+
+template <index_t MPerBlock, index_t NPerBlock>
+struct BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
 
     __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt() = default;
 
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const BlockToCTileMap_M00_N0_M01Adapt&) =
+        default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(BlockToCTileMap_M00_N0_M01Adapt&&) =
+        default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt&
+    operator=(const BlockToCTileMap_M00_N0_M01Adapt&) = default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt&
+    operator=(BlockToCTileMap_M00_N0_M01Adapt&&) = default;
+
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(index_t M, index_t N, index_t M01 = 8)
+        : M_(M), N_(N), M01_(M01)
+    {
+    }
+
+    template <typename CGridDesc_M_N>
     __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const CGridDesc_M_N& c_grid_desc_m_n,
                                                         index_t M01 = 8)
-        : M01_(M01), c_grid_desc_m_n_(c_grid_desc_m_n)
+        : BlockToCTileMap_M00_N0_M01Adapt(
+              c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1), M01)
     {
     }
 
-    __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+    __host__ static constexpr index_t CalculateGridSize(index_t M, index_t N)
     {
-        const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I0), MPerBlock);
-        const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I1), NPerBlock);
+        const auto M0 = math::integer_divide_ceil(M, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N, NPerBlock);
 
-        const index_t grid_size = M0 * N0;
+        return M0 * N0;
+    }
 
-        return grid_size;
+    template <typename CGridDesc_M_N>
+    __host__ static constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        return CalculateGridSize(c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1));
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const
+    {
+        return true;
     }
 
     template <typename TopIdx>
@@ -140,8 +167,8 @@ struct BlockToCTileMap_M00_N0_M01Adapt
     {
         auto block_1d_id = idx_top[I0];
 
-        const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n_.GetLength(I0), MPerBlock);
-        const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n_.GetLength(I1), NPerBlock);
+        const auto M0 = math::integer_divide_ceil(M_, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N_, NPerBlock);
 
         block_1d_id = block_1d_id % (M0 * N0); // swallow batch index
 
@@ -154,6 +181,50 @@ struct BlockToCTileMap_M00_N0_M01Adapt
         index_t idx_M01          = idx_M0 % M01_;
         index_t idx_N0_M01_local = idx_N0 + idx_M01 * N0;
 
+        /**
+         *                        idxN0
+         *
+         *           |<               mtx   N                 >|
+         *
+         *             NPerBlock   NPerBlock   NPerBlock   NPerBlock
+         *                N_0         N_1        N_2         N_3
+         *       -   |-----------|-----------|-----------|-----|-----|-
+         *       ^   | -   -  0  |/---->  2  |           |     |     |
+         *           | |   |     /     |     |           |     |     |  M_0  MPerBlock
+         *           | M   |    /|     |     |           |     |     |
+         *           |-0---|---/-|-----|-----|-----------|-----|-----|-
+         *           | 1   |  /  |     |     |  blockid  |     |     |
+         * idxM0     | |   | /   |     V     |     5     |     |     |  M_1  MPerBlock
+         *           | -   V   1 |     -  3  |           |     |     |
+         *           |-----------|-----------|-----------|-----|-----|-
+         *    mtx M  |           |           |           |     |     |
+         *           |           |           |           |     |     |  M_2  MPerBlock
+         *           |           |           |           |     |     |
+         *           |-----------|-----------|-----------|-----|-----|-
+         *           |           |           |           |     |     |
+         *           |           |           |           |     |     |  M_3  MPerBlock
+         *           |           |           |           |     |     |
+         *           |-----------|-----------|-----------|-----|-----|-
+         *       V   |           |           |           |     |     |
+         *       -   |-----------|-----------|-----------|-----|-----|- M_4  MPerBlock
+         *           |           |           |           |     |     |
+         *           |-----------|-----------|-----------|-----|-----|-
+         *  Example:
+         *   assume:
+         *      M0 = 5
+         *      N0 = 4
+         *      block_1d_id = 5
+         *      M01 = 2
+         *
+         *   idx_N0 = 1
+         *   idx_M0 = 1
+         *   M01_adapt = 2
+         *   idx_M00 = 0
+         *   idx_M01 = 1
+         *   idx_N0_M01_local = 5
+         *   output {1, 2}
+         */
+
         return make_tuple(idx_N0_M01_local % M01_adapt + idx_M00 * M01_,
                           idx_N0_M01_local / M01_adapt);
     }
@@ -165,11 +236,18 @@ struct BlockToCTileMap_M00_N0_M01Adapt
         return true; // always valid provided that user gets grid size from CalculateGridSize()
     }
 
-    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const { return true; }
-
     private:
+    index_t M_;
+    index_t N_;
     index_t M01_;
-    CGridDesc_M_N c_grid_desc_m_n_;
+};
+
+// keep the redundant type argument for backward compatibility
+template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N>
+struct BlockToCTileMap_M00_N0_M01Adapt : BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
+{
+    using BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>::
+        BlockToCTileMap_M00_N0_M01Adapt;
 };
 
 // 2D slices of column-vectors in 3D space
@@ -543,4 +621,52 @@ struct OffsettedBlockToCTileMap
     index_t block_start_;
 };
 
+/**
+ * @brief      Simple tile mapping which creates 3D grid of block of threads.
+ *
+ * @paragraph  Description
+ *             This Block-to-C-tile-map creates a 3D grid (n_blocks, m_blocks, z_blocks) of thread
+ *             blocks. The first 2D are regular 2D tiles created by division of output GEMM
+ *             dimenions by corresponding tile size. The third dimension (Z) is a k-split dimension,
+ *             which denotes the number of blocks we use to divide work on GEMM K dimension onto.
+ *
+ * @tparam     MPerBlock  Output block tile size in M dimension.
+ * @tparam     NPerBlock  Output block tile size in N dimension.
+ */
+template <index_t MPerBlock, index_t NPerBlock>
+struct BlockToCTileMap_3DGrid_KSplit
+{
+
+    __host__ __device__ BlockToCTileMap_3DGrid_KSplit() = default;
+
+    __host__ __device__ constexpr auto
+    CalculateGridSize(index_t M, index_t N, index_t k_split) const
+    {
+        // Create 3D grid
+        const auto M0 = math::integer_divide_ceil(M, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N, NPerBlock);
+
+        return std::make_tuple(N0, M0, k_split);
+    }
+
+    template <typename TopIdx>
+    __device__ constexpr auto CalculateBottomIndex(const TopIdx&) const
+    {
+        return make_tuple(blockIdx.z, blockIdx.y, blockIdx.x);
+    }
+
+    template <typename CTileIdx, typename CTileDim>
+    __host__ __device__ bool ValidCTileIndex(const CTileIdx& /* c_tile_idx */,
+                                             const CTileDim& /* c_tile_dim */) const
+    {
+        return true; // always valid provided that user gets grid size from CalculateGridSize()
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const
+    {
+        return true;
+    }
+};
+
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..523e7f7c539da7ac6e5729c6e87b4a21f1bd3f6b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
@@ -0,0 +1,1111 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+
+namespace ck {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   output : F[M, N0], where N0 is number of blocks along N dimension
+//   output : G[M, N0], where N0 is number of blocks along N dimension
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+//   F, G = welford(E)
+// Assume:
+//   D0, D1, ... and E have the same layout
+//   Calculate mean & variance along N dimension for E
+template <typename ABDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EMeanVarDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+          typename AGridDesc_M_K,
+          typename BGridDesc_N_K,
+          typename DsGridDesc_M_N,
+          typename EGridDesc_M_N,
+          typename MeanVarGridDesc_M_NBlock,
+          typename CountGridDesc_M_NBlock,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1Value,
+          index_t BK1Value,
+          index_t MPerXdl,
+          index_t NPerXdl,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename PostShuffleThreadClusterSize_M_N,
+          index_t PostShuffleScalarPerVector,
+          LoopScheduler LoopSched,
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct GridwiseGemmMultipleDWelfordFirstHalf_xdl_cshuffle
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto AK1         = Number<AK1Value>{};
+    static constexpr auto BK1         = Number<BK1Value>{};
+    static constexpr auto AK0PerBlock = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0PerBlock = Number<KPerBlock / BK1Value>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    using GridwiseGemmPipe = remove_cvref_t<decltype(
+        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(AK0PerBlock, Number<MPerBlock>{}, AK1),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(BK0PerBlock, Number<NPerBlock>{}, BK1),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    {
+        constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+        constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl>{},
+                           I1,
+                           Number<CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>{}));
+
+        return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // ck::Tuple<const D0DataType*, const D1DataType*, ...>
+    static constexpr auto MakeDsGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                return static_cast<const DDataType*>(nullptr);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+
+        // LDS allocation for C shuffle in LDS
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+        constexpr auto c_block_size =
+            c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
+
+        return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
+                             sizeof(ABDataType),
+                         c_block_size * sizeof(CShuffleDataType));
+    }
+
+    // A desc for source in blockwise copy
+    __host__ __device__ static constexpr auto
+    MakeDefaultAGridDescriptor_AK0_M_AK1(const AGridDesc_M_K& a_grid_desc_m_k)
+    {
+        const auto M = a_grid_desc_m_k.GetLength(I0);
+        const auto K = a_grid_desc_m_k.GetLength(I1);
+
+        const auto AK0 = K / AK1;
+
+        return transform_tensor_descriptor(a_grid_desc_m_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
+                                                      make_pass_through_transform(M)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    // B desc for source in blockwise copy
+    __host__ __device__ static constexpr auto
+    MakeDefaultBGridDescriptor_BK0_N_BK1(const BGridDesc_N_K& b_grid_desc_n_k)
+    {
+        const auto N = b_grid_desc_n_k.GetLength(I0);
+        const auto K = b_grid_desc_n_k.GetLength(I1);
+
+        const auto BK0 = K / BK1;
+
+        return transform_tensor_descriptor(b_grid_desc_n_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                                                      make_pass_through_transform(N)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    // E desc for destination in blockwise copy
+    template <typename EGridDescriptor_M_N>
+    __host__ __device__ static constexpr auto MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+        const EGridDescriptor_M_N& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            e_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return e_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // Ds desc for source in blockwise copy
+    template <typename DsGridDescriptor_M_N>
+    __host__ __device__ static constexpr auto
+    MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+        const DsGridDescriptor_M_N& ds_grid_desc_m_n)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    template <typename GridDescriptor_M_N>
+    __host__ __device__ static constexpr auto
+    MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(const GridDescriptor_M_N& grid_desc_m_n)
+    {
+        const auto M      = grid_desc_m_n.GetLength(I0);
+        const auto NBlock = grid_desc_m_n.GetLength(I1);
+        const auto MBlock = M / MPerBlock;
+
+        const auto grid_desc_mblock_mperblock_nblock = transform_tensor_descriptor(
+            grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_pass_through_transform(NBlock)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2>{}));
+
+        return grid_desc_mblock_mperblock_nblock;
+    }
+
+    // return block_id to E matrix tile idx (m0, n0) mapping
+    __host__ __device__ static constexpr auto
+    MakeDefaultBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, EGridDesc_M_N>(
+            e_grid_desc_m_n);
+    }
+
+    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
+    template <typename Block2ETileMap>
+    __host__ __device__ static constexpr bool CheckValidity(const AGridDesc_M_K& a_grid_desc_m_k,
+                                                            const BGridDesc_N_K& b_grid_desc_n_k,
+                                                            const DsGridDesc_M_N& ds_grid_desc_m_n,
+                                                            const EGridDesc_M_N& e_grid_desc_m_n,
+                                                            const Block2ETileMap& block_2_etile_map)
+    {
+        static_assert((MPerBlock % (MPerXdl * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
+                      "Invalid tuning param!");
+
+        const auto M = a_grid_desc_m_k.GetLength(I0);
+        const auto N = b_grid_desc_n_k.GetLength(I0);
+        const auto K = a_grid_desc_m_k.GetLength(I1);
+
+        // check consistency of desc
+        if(!(M == e_grid_desc_m_n.GetLength(I0) && N == e_grid_desc_m_n.GetLength(I1)))
+        {
+            return false;
+        }
+
+        bool valid = true;
+
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            valid = valid && (M == ds_grid_desc_m_n[i].GetLength(I0) &&
+                              N == ds_grid_desc_m_n[i].GetLength(I1));
+        });
+
+        if(!valid)
+        {
+            return false;
+        }
+
+        // check tile size
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0))
+        {
+            return false;
+        }
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = K / KPerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        // check block-to-E-tile
+        if(!block_2_etile_map.CheckValidity(e_grid_desc_m_n))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        // check tensor size: cannot be larger than 2GB each
+        constexpr long_index_t TwoGB = (long_index_t{1} << 31);
+
+        if(!(a_grid_desc_m_k.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+             b_grid_desc_n_k.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+             e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EMeanVarDataType) <= TwoGB))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / KPerBlock;
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    using DefaultAGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
+    using DefaultBGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
+        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using MeanVarGridDescriptor_MBlock_MPerBlock_NBlock      = remove_cvref_t<decltype(
+        MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(MeanVarGridDesc_M_NBlock{}))>;
+    using CountGridDescriptor_MBlock_MPerBlock_NBlock        = remove_cvref_t<decltype(
+        MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(CountGridDesc_M_NBlock{}))>;
+    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
+        MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
+
+    using DefaultBlock2ETileMap =
+        remove_cvref_t<decltype(MakeDefaultBlock2ETileMap(EGridDesc_M_N{}))>;
+
+    using DsGridPointer = decltype(MakeDsGridPointer());
+
+    template <bool HasMainKBlockLoop,
+              typename AGridDesc_AK0_M_AK1,
+              typename BGridDesc_BK0_N_BK1,
+              typename Block2ETileMap>
+    __device__ static void
+    Run(const ABDataType* __restrict__ p_a_grid,
+        const ABDataType* __restrict__ p_b_grid,
+        DsGridPointer p_ds_grid,
+        EMeanVarDataType* __restrict__ p_e_grid,
+        EMeanVarDataType* __restrict__ p_welford_mean_grid,
+        EMeanVarDataType* __restrict__ p_welford_var_grid,
+        int32_t* __restrict__ p_welford_count,
+        void* __restrict__ p_shared,
+        const AElementwiseOperation& a_element_op,
+        const BElementwiseOperation& b_element_op,
+        const CDEElementwiseOperation& cde_element_op,
+        const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
+        const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
+        const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
+            ds_grid_desc_mblock_mperblock_nblock_nperblock,
+        const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
+            e_grid_desc_mblock_mperblock_nblock_nperblock,
+        const MeanVarGridDescriptor_MBlock_MPerBlock_NBlock&
+            mean_var_grid_desc_mblock_mperblock_nblock,
+        const CountGridDescriptor_MBlock_MPerBlock_NBlock& count_grid_desc_mblock_mperblock_nblock,
+        const Block2ETileMap& block_2_etile_map,
+        index_t NRaw)
+    {
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        const auto ds_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_ds_grid[i],
+                    ds_grid_desc_mblock_mperblock_nblock_nperblock[i].GetElementSpaceSize());
+            },
+            Number<NumDTensor>{});
+
+        auto e_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_e_grid, e_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+        auto mean_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_welford_mean_grid, mean_var_grid_desc_mblock_mperblock_nblock.GetElementSpaceSize());
+
+        auto var_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_welford_var_grid, mean_var_grid_desc_mblock_mperblock_nblock.GetElementSpaceSize());
+
+        auto welford_count_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_welford_count, count_grid_desc_mblock_mperblock_nblock.GetElementSpaceSize());
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            block_2_etile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        if(!block_2_etile_map.ValidCTileIndex(
+               block_work_idx,
+               make_tuple(e_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
+                          e_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
+        {
+            return;
+        }
+
+        // HACK: this force m/n_block_data_idx_on_grid into SGPR
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+
+        const index_t n_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // A matrix blockwise copy
+        auto a_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                AElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<AK0PerBlock, MPerBlock, AK1>,
+                                                ABlockTransferThreadClusterLengths_AK0_M_AK1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                ABDataType,
+                                                ABDataType,
+                                                decltype(a_grid_desc_ak0_m_ak1),
+                                                decltype(a_block_desc_ak0_m_ak1),
+                                                ABlockTransferSrcAccessOrder,
+                                                Sequence<1, 0, 2>,
+                                                ABlockTransferSrcVectorDim,
+                                                2,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_AK1,
+                                                1,
+                                                1,
+                                                AThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                a_grid_desc_ak0_m_ak1,
+                make_multi_index(0, m_block_data_idx_on_grid, 0),
+                a_element_op,
+                a_block_desc_ak0_m_ak1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // B matrix blockwise copy
+        auto b_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<BK0PerBlock, NPerBlock, BK1>,
+                                                BBlockTransferThreadClusterLengths_BK0_N_BK1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                ABDataType,
+                                                ABDataType,
+                                                decltype(b_grid_desc_bk0_n_bk1),
+                                                decltype(b_block_desc_bk0_n_bk1),
+                                                BBlockTransferSrcAccessOrder,
+                                                Sequence<1, 0, 2>,
+                                                BBlockTransferSrcVectorDim,
+                                                2,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_BK1,
+                                                1,
+                                                1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                b_grid_desc_bk0_n_bk1,
+                make_multi_index(0, n_block_data_idx_on_grid, 0),
+                b_element_op,
+                b_block_desc_bk0_n_bk1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // GEMM definition
+        //   c_mtx += transpose(a_mtx) * b_mtx
+        //     a_mtx[K0PerBlock, MPerBlock] is in LDS
+        //     b_mtx[K0PerBlock, NPerBlock] is in LDS
+        //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
+        //       register
+        // sanity check
+        constexpr index_t KPack =
+            math::max(math::lcm(AK1, BK1),
+                      MfmaSelector<ABDataType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+
+        auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
+            BlockSize,
+            ABDataType,
+            AccDataType,
+            decltype(a_block_desc_ak0_m_ak1),
+            decltype(b_block_desc_bk0_n_bk1),
+            MPerXdl,
+            NPerXdl,
+            MXdlPerWave,
+            NXdlPerWave,
+            KPack,
+            LoopSched>();
+
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataType*>(p_shared) + a_block_space_size_aligned,
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1, 0, 0);
+
+        // gridwise GEMM pipeline
+        const auto gridwise_gemm_pipeline =
+            GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>();
+
+        const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
+            (a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2)) /
+            KPerBlock);
+
+        gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_grid_desc_ak0_m_ak1,
+                                                               a_block_desc_ak0_m_ak1,
+                                                               a_blockwise_copy,
+                                                               a_grid_buf,
+                                                               a_block_buf,
+                                                               a_block_slice_copy_step,
+                                                               b_grid_desc_bk0_n_bk1,
+                                                               b_block_desc_bk0_n_bk1,
+                                                               b_blockwise_copy,
+                                                               b_grid_buf,
+                                                               b_block_buf,
+                                                               b_block_slice_copy_step,
+                                                               blockwise_gemm,
+                                                               c_thread_buf,
+                                                               num_k_block_main_loop);
+
+        // shuffle C, Welford and write out
+        {
+            static_assert(MXdlPerWave % CShuffleMXdlPerWavePerShuffle == 0 &&
+                              NXdlPerWave % CShuffleNXdlPerWavePerShuffle == 0,
+                          "wrong!");
+
+            constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+            constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+            // TODO: hacky, fix it!
+            constexpr auto c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2 =
+                blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            // TODO: hacky, fix it!
+            // c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp is only used to get lengths
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp =
+                blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            constexpr auto M0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I0);
+            constexpr auto N0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I1);
+            constexpr auto M1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I2);
+            constexpr auto N1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I3);
+            constexpr auto M2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I4);
+            constexpr auto M3 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I5);
+            constexpr auto M4 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I6);
+            constexpr auto N2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I7);
+
+            constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+                GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+            auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                static_cast<CShuffleDataType*>(p_shared),
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2 = transform_tensor_descriptor(
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleMXdlPerWavePerShuffle>{}, // M0 (MXdlPerWave) per shuffle
+                        M1,                                      // M1 = MWave
+                        M2,                                      // M2 * M3 * M4 = MPerXdl
+                        M3,
+                        M4)),
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleNXdlPerWavePerShuffle>{}, // N0 (NXdlPerWave) per shuffle
+                        N1,                                      // N1 = NWave
+                        N2))),                                   // N2 = NPerXdl
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(
+                    Sequence<>{}, Sequence<0, 2, 4, 5, 6>{}, Sequence<>{}, Sequence<1, 3, 7>{}));
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block =
+                blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
+
+            const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+            const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(M0, M1, M2, M3, M4))),
+                    make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto m_thread_data_on_block_idx =
+                m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor.CalculateBottomIndex(
+                    make_multi_index(m_thread_data_on_block));
+
+            const auto n_thread_data_on_block_to_n0_n1_n2_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
+                    make_tuple(Sequence<0, 1, 2>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_block_idx =
+                n_thread_data_on_block_to_n0_n1_n2_adaptor.CalculateBottomIndex(
+                    make_multi_index(n_thread_data_on_block));
+
+            // shuffle: threadwise copy C from VGPR to LDS
+            auto c_thread_copy_vgpr_to_lds =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   CShuffleDataType,
+                                                   decltype(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   decltype(c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Sequence<CShuffleMXdlPerWavePerShuffle,
+                                                            CShuffleNXdlPerWavePerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            M2,
+                                                            I1,
+                                                            M4,
+                                                            I1>,
+                                                   Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                                   7,
+                                                   1,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>{
+                    c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                    make_multi_index(0,
+                                     0,
+                                     m_thread_data_on_block_idx[I1],
+                                     n_thread_data_on_block_idx[I1],
+                                     m_thread_data_on_block_idx[I2],
+                                     m_thread_data_on_block_idx[I3],
+                                     m_thread_data_on_block_idx[I4],
+                                     n_thread_data_on_block_idx[I2]),
+                    ck::tensor_operation::element_wise::PassThrough{}};
+
+            // space filling curve for threadwise C in VGPR
+            constexpr auto sfc_c_vgpr =
+                SpaceFillingCurve<Sequence<MXdlPerWave, NXdlPerWave, 1, 1, M2, 1, M4, 1>,
+                                  Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                  Sequence<CShuffleMXdlPerWavePerShuffle,
+                                           CShuffleNXdlPerWavePerShuffle,
+                                           1,
+                                           1,
+                                           M2,
+                                           1,
+                                           M4,
+                                           1>,
+                                  false>{};
+
+            // space filling curve for shuffled blockwise C in global mem
+            constexpr auto sfc_der_global =
+                SpaceFillingCurve<Sequence<1, MPerBlock, 1, NPerBlock>,
+                                  Sequence<0, 2, 1, 3>,
+                                  Sequence<1,
+                                           CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                                           1,
+                                           CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>,
+                                  false>{};
+
+            // LDS c_shuffle_block_desc_mperblock_nperblock
+            constexpr auto c_shuffle_block_desc_mperblock_nperblock = transform_tensor_descriptor(
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_pass_through_transform(
+                        c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetLength(I1)),
+                    make_freeze_transform(I0),
+                    make_pass_through_transform(
+                        c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetLength(I3))),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<>{}, Sequence<0>{}, Sequence<>{}, Sequence<1>{}));
+
+            static_assert(PostShuffleThreadClusterSize_M_N::At(I0) *
+                                  PostShuffleThreadClusterSize_M_N::At(I1) ==
+                              BlockSize,
+                          "wrong!");
+
+            static_assert((CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl) %
+                                      PostShuffleThreadClusterSize_M_N::At(I0) ==
+                                  0 &&
+                              (CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl) %
+                                      PostShuffleThreadClusterSize_M_N::At(I1) ==
+                                  0,
+                          "wrong!");
+
+            constexpr index_t PostShuffleThreadSliceSize_M =
+                (CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl) /
+                PostShuffleThreadClusterSize_M_N::At(I0);
+
+            constexpr index_t PostShuffleThreadSliceSize_N =
+                (CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl) /
+                PostShuffleThreadClusterSize_M_N::At(I1);
+
+            constexpr auto PostShuffleThreadSliceSize_M_N =
+                Sequence<PostShuffleThreadSliceSize_M, PostShuffleThreadSliceSize_N>{};
+
+            // VGPR post_shuffle_thread_desc_m_n
+            constexpr auto post_shuffle_thread_desc_m_n = make_naive_tensor_descriptor_packed(
+                make_tuple(Number<PostShuffleThreadSliceSize_M>{},
+                           Number<PostShuffleThreadSliceSize_N>{}));
+
+            auto e_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, AccDataType>(
+                post_shuffle_thread_desc_m_n.GetElementSpaceSize());
+
+            // To apply D0, D1, ... and Welford.
+            // threadwise copy from LDS to VGPR
+            constexpr auto post_shuffle_thread_cluster_desc =
+                make_cluster_descriptor(PostShuffleThreadClusterSize_M_N{}, Sequence<0, 1>{});
+
+            const auto post_shuffle_thread_cluster_idx =
+                post_shuffle_thread_cluster_desc.CalculateBottomIndex(
+                    make_multi_index(get_thread_local_1d_id()));
+
+            const auto post_shuffle_thread_data_idx_begin =
+                post_shuffle_thread_cluster_idx * PostShuffleThreadSliceSize_M_N;
+
+            // To apply D0, D1, ... and Welford.
+            // Copy c shuffle from LDS back to VGPR
+            auto post_shuffle_thread_copy_lds_to_vgpr =
+                ThreadwiseTensorSliceTransfer_v2<CShuffleDataType,
+                                                 AccDataType,
+                                                 decltype(c_shuffle_block_desc_mperblock_nperblock),
+                                                 decltype(post_shuffle_thread_desc_m_n),
+                                                 decltype(PostShuffleThreadSliceSize_M_N),
+                                                 Sequence<0, 1>,
+                                                 1,
+                                                 PostShuffleScalarPerVector,
+                                                 1,
+                                                 true>{c_shuffle_block_desc_mperblock_nperblock,
+                                                       post_shuffle_thread_data_idx_begin};
+
+            // D0, D1, ..., Dn
+            constexpr auto post_shuffle_thread_desc_I1_mperblock_I1_nperblock =
+                make_naive_tensor_descriptor_packed(
+                    make_tuple(I1,
+                               Number<PostShuffleThreadSliceSize_M>{},
+                               I1,
+                               Number<PostShuffleThreadSliceSize_N>{}));
+
+            // FIXME: Decrease usage of VGPR
+            // Apply pointwise lambda function from multi-source (Global and LDS) into VGPR
+            auto ds_thread_buf = generate_tuple(
+                [&](auto) {
+                    return make_static_buffer<AddressSpaceEnum::Vgpr, CShuffleDataType>(
+                        post_shuffle_thread_desc_I1_mperblock_I1_nperblock.GetElementSpaceSize());
+                },
+                Number<NumDTensor>{});
+
+            // Copy D0, D1, ..., Dn from global to VGPR
+            auto ds_thread_copy_global_to_vgpr = generate_tuple(
+                [&](auto I) {
+                    using DDataType = remove_cvref_t<tuple_element_t<I.value, DsDataType>>;
+                    return ThreadwiseTensorSliceTransfer_v2<
+                        DDataType,
+                        AccDataType,
+                        decltype(ds_grid_desc_mblock_mperblock_nblock_nperblock[I]),
+                        decltype(post_shuffle_thread_desc_I1_mperblock_I1_nperblock),
+                        Sequence<I1,
+                                 PostShuffleThreadSliceSize_M,
+                                 I1,
+                                 PostShuffleThreadSliceSize_N>,
+                        Sequence<0, 1, 2, 3>,
+                        3,
+                        PostShuffleScalarPerVector,
+                        1,
+                        true>(
+                        ds_grid_desc_mblock_mperblock_nblock_nperblock[I],
+                        make_multi_index(
+                            I0,
+                            m_block_data_idx_on_grid + post_shuffle_thread_data_idx_begin[I0],
+                            I0,
+                            n_block_data_idx_on_grid + post_shuffle_thread_data_idx_begin[I1]));
+                },
+                Number<NumDTensor>{});
+
+            auto e_thread_copy_vgpr_to_global = ThreadwiseTensorSliceTransfer_v1r3<
+                AccDataType,
+                EMeanVarDataType,
+                decltype(post_shuffle_thread_desc_I1_mperblock_I1_nperblock),
+                decltype(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                tensor_operation::element_wise::PassThrough,
+                Sequence<I1,
+                         PostShuffleThreadSliceSize_M,
+                         I1,
+                         PostShuffleThreadSliceSize_N>, // SliceLengths
+                Sequence<0, 1, 2, 3>,                   // DimAccessOrder
+                3,                                      // DstVectorDim
+                PostShuffleScalarPerVector,
+                InMemoryDataOperationEnum::Set,
+                1,
+                true>{
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+                make_multi_index(I0,
+                                 m_block_data_idx_on_grid + post_shuffle_thread_data_idx_begin[I0],
+                                 I0,
+                                 n_block_data_idx_on_grid + post_shuffle_thread_data_idx_begin[I1]),
+                tensor_operation::element_wise::PassThrough{}};
+
+            // Welford
+            constexpr auto thread_welford_src_desc_m_k = make_naive_tensor_descriptor_packed(
+                make_tuple(Number<PostShuffleThreadSliceSize_M>{},
+                           Number<PostShuffleThreadSliceSize_N>{}));
+
+            constexpr auto thread_welford_dst_desc_m = make_naive_tensor_descriptor_packed(
+                make_tuple(Number<PostShuffleThreadSliceSize_M>{}));
+
+            using ThreadwiseWelford = ThreadwiseWelford<AccDataType,
+                                                        decltype(thread_welford_src_desc_m_k),
+                                                        decltype(thread_welford_dst_desc_m)>;
+
+            using BlockwiseWelford = BlockwiseWelford<AccDataType,
+                                                      BlockSize,
+                                                      PostShuffleThreadClusterSize_M_N,
+                                                      Sequence<0, 1>,
+                                                      false>;
+
+            constexpr int num_shuffleM =
+                MPerBlock / (CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl);
+
+            constexpr int num_shuffleN =
+                NPerBlock / (CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl);
+
+            using mean_var_vgpr_type =
+                decltype(make_static_buffer<AddressSpaceEnum::Vgpr, AccDataType>(
+                    thread_welford_dst_desc_m.GetElementSpaceSize()));
+
+            using welford_count_vgpr_type =
+                decltype(make_static_buffer<AddressSpaceEnum::Vgpr, int32_t>(
+                    thread_welford_dst_desc_m.GetElementSpaceSize()));
+
+            Array<ThreadwiseWelford, num_shuffleM> threadwise_welfords;
+            Array<mean_var_vgpr_type, num_shuffleM> mean_thread_bufs;
+            Array<mean_var_vgpr_type, num_shuffleM> var_thread_bufs;
+            Array<welford_count_vgpr_type, num_shuffleM> welford_count_thread_bufs;
+
+            int max_count     = PostShuffleThreadSliceSize_N * num_shuffleN;
+            const auto nblock = mean_var_grid_desc_mblock_mperblock_nblock.GetLength(I2);
+
+            // tail block
+            if(block_work_idx[I1] % nblock == nblock - 1)
+            {
+                constexpr index_t NPerShuffleBlock =
+                    CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl;
+
+                int NPerBlockTail = NRaw - NPerBlock * (nblock - 1);
+                int thread_max_len =
+                    PostShuffleThreadSliceSize_N * (post_shuffle_thread_cluster_idx[I1] + 1);
+                int shuffle_step = 0;
+                while(thread_max_len <= NPerBlockTail && shuffle_step < num_shuffleN)
+                {
+                    ++shuffle_step;
+                    thread_max_len += NPerShuffleBlock;
+                }
+
+                int delta = 0;
+                if(thread_max_len - NPerBlockTail > PostShuffleThreadSliceSize_N)
+                    delta = 0;
+                else if(NPerBlockTail > thread_max_len)
+                    delta = PostShuffleThreadSliceSize_N;
+                else
+                    delta = PostShuffleThreadSliceSize_N - thread_max_len + NPerBlockTail;
+
+                max_count = shuffle_step * PostShuffleThreadSliceSize_N + delta;
+            }
+
+            static_for<0, num_shuffleM, 1>{}([&](auto i) {
+                threadwise_welfords(i).max_count_ = max_count;
+                mean_thread_bufs(i) = make_static_buffer<AddressSpaceEnum::Vgpr, AccDataType>(
+                    thread_welford_dst_desc_m.GetElementSpaceSize());
+
+                var_thread_bufs(i) = make_static_buffer<AddressSpaceEnum::Vgpr, AccDataType>(
+                    thread_welford_dst_desc_m.GetElementSpaceSize());
+
+                welford_count_thread_bufs(i) = make_static_buffer<AddressSpaceEnum::Vgpr, int32_t>(
+                    thread_welford_dst_desc_m.GetElementSpaceSize());
+
+                static_for<0, PostShuffleThreadSliceSize_M, 1>{}([&](auto j) {
+                    mean_thread_bufs(i)(j)          = type_convert<AccDataType>(0.0f);
+                    var_thread_bufs(i)(j)           = type_convert<AccDataType>(0.0f);
+                    welford_count_thread_bufs(i)(j) = 0;
+                });
+            });
+
+            constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+            static_assert(num_access == sfc_der_global.GetNumOfAccess(), "wrong!");
+
+            int shuffleM_index = __builtin_amdgcn_readfirstlane(0);
+            static_for<0, num_access, 1>{}([&](auto access_id) {
+                // make sure it's safe to read from LDS
+                block_sync_lds();
+
+                // each thread shuffle data from VGPR to LDS
+                c_thread_copy_vgpr_to_lds.Run(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                              c_thread_buf,
+                                              c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              c_shuffle_block_buf);
+
+                // make sure it's safe to write to LDS
+                block_sync_lds();
+
+                // Get shuffle data from LDS to VGPR
+                post_shuffle_thread_copy_lds_to_vgpr.Run(c_shuffle_block_desc_mperblock_nperblock,
+                                                         c_shuffle_block_buf,
+                                                         post_shuffle_thread_desc_m_n,
+                                                         make_tuple(I0, I0),
+                                                         e_thread_buf);
+
+                // Global read D0, D1, ...
+                static_for<0, NumDTensor, 1>{}([&](auto Id) {
+                    auto& d_thread_copy_global_to_vgpr = ds_thread_copy_global_to_vgpr(Id);
+                    d_thread_copy_global_to_vgpr.Run(
+                        ds_grid_desc_mblock_mperblock_nblock_nperblock[Id],
+                        ds_grid_buf[Id],
+                        post_shuffle_thread_desc_I1_mperblock_I1_nperblock,
+                        make_tuple(I0, I0, I0, I0),
+                        ds_thread_buf(Id));
+
+                    if constexpr(access_id < num_access - 1)
+                    {
+                        // move on D0, D1, ...
+                        constexpr auto de_global_step = sfc_der_global.GetForwardStep(access_id);
+                        d_thread_copy_global_to_vgpr.MoveSrcSliceWindow(
+                            ds_grid_desc_mblock_mperblock_nblock_nperblock[Id], de_global_step);
+                    }
+                });
+
+                // cde_element_op(e, c, d0, d1, ...);
+                static_for<0, post_shuffle_thread_desc_m_n.GetElementSize(), 1>{}([&](auto i) {
+                    const auto c_ds_src_data_refs = concat_tuple_of_reference(
+                        tie(e_thread_buf[i]),
+                        generate_tie(
+                            [&](auto Id) -> const auto& { return ds_thread_buf[Id][i]; },
+                            Number<NumDTensor>{}));
+                    auto e_dst_data_refs = tie(e_thread_buf(i));
+                    unpack2(cde_element_op, e_dst_data_refs, c_ds_src_data_refs);
+                });
+
+                // Global write E
+                e_thread_copy_vgpr_to_global.Run(post_shuffle_thread_desc_I1_mperblock_I1_nperblock,
+                                                 make_tuple(I0, I0, I0, I0),
+                                                 e_thread_buf,
+                                                 e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                 e_grid_buf);
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    // move on E
+                    constexpr auto de_global_step = sfc_der_global.GetForwardStep(access_id);
+                    e_thread_copy_vgpr_to_global.MoveDstSliceWindow(
+                        e_grid_desc_mblock_mperblock_nblock_nperblock, de_global_step);
+                }
+
+                // Threadwise welford
+                auto& threadwise_welford = threadwise_welfords(shuffleM_index);
+                auto& mean_thread_buf    = mean_thread_bufs(shuffleM_index);
+                auto& var_thread_buf     = var_thread_bufs(shuffleM_index);
+
+                threadwise_welford.Run(e_thread_buf, mean_thread_buf, var_thread_buf);
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    constexpr auto de_global_step = sfc_der_global.GetForwardStep(access_id);
+                    constexpr int shuffleMInc =
+                        de_global_step[I1] /
+                        c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetLength(I1);
+                    shuffleM_index = __builtin_amdgcn_readfirstlane(shuffleM_index + shuffleMInc);
+                }
+            }); // copy c, d, e + welford
+
+            // Blockwise welford and write out
+            static_for<0, num_shuffleM, 1>{}([&](auto i) {
+                auto& mean_thread_buf  = mean_thread_bufs(i);
+                auto& var_thread_buf   = var_thread_bufs(i);
+                auto& count_thread_buf = welford_count_thread_bufs(i);
+
+                static_for<0, PostShuffleThreadSliceSize_M, 1>{}([&](auto j) {
+                    block_sync_lds();
+                    count_thread_buf(j) = threadwise_welfords(i).cur_count_;
+                    BlockwiseWelford::Run(
+                        mean_thread_buf(j), var_thread_buf(j), count_thread_buf(j));
+                });
+
+                if(post_shuffle_thread_cluster_idx[I1] == 0)
+                {
+                    constexpr auto thread_welford_desc_I_m_I = make_naive_tensor_descriptor_packed(
+                        make_tuple(I1, Number<PostShuffleThreadSliceSize_M>{}, I1));
+
+                    constexpr int shuffleMPerBlock =
+                        c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetLength(I1);
+
+                    auto mean_var_count_thread_copy_index = make_multi_index(
+                        block_work_idx[I0],                                            // mblock
+                        shuffleMPerBlock * i + post_shuffle_thread_data_idx_begin[I0], // mperblock
+                        block_work_idx[I1]);                                           // nblock
+
+                    auto mean_var_thread_copy_vgpr_to_global = ThreadwiseTensorSliceTransfer_v1r3<
+                        AccDataType,
+                        EMeanVarDataType,
+                        decltype(thread_welford_desc_I_m_I),
+                        decltype(mean_var_grid_desc_mblock_mperblock_nblock),
+                        tensor_operation::element_wise::PassThrough,
+                        Sequence<1, PostShuffleThreadSliceSize_M, 1>,
+                        Sequence<0, 1, 2>,
+                        1,
+                        1,
+                        InMemoryDataOperationEnum::Set,
+                        1,
+                        true>{mean_var_grid_desc_mblock_mperblock_nblock,
+                              mean_var_count_thread_copy_index,
+                              tensor_operation::element_wise::PassThrough{}};
+
+                    mean_var_thread_copy_vgpr_to_global.Run(
+                        thread_welford_desc_I_m_I,
+                        make_tuple(I0, I0, I0),
+                        mean_thread_buf,
+                        mean_var_grid_desc_mblock_mperblock_nblock,
+                        mean_grid_buf); // write mean
+
+                    mean_var_thread_copy_vgpr_to_global.Run(
+                        thread_welford_desc_I_m_I,
+                        make_tuple(I0, I0, I0),
+                        var_thread_buf,
+                        mean_var_grid_desc_mblock_mperblock_nblock,
+                        var_grid_buf); // write variance
+
+                    // Stride of count is [0, 1]. Only the first row in count[0, 0:nblock] need
+                    // to be written.
+                    if(i == 0 && block_work_idx[I0] == 0 &&
+                       post_shuffle_thread_cluster_idx[I0] == 0)
+                    {
+                        auto count_thread_copy_vgpr_to_global = ThreadwiseTensorSliceTransfer_v1r3<
+                            int32_t,
+                            int32_t,
+                            decltype(thread_welford_desc_I_m_I),
+                            decltype(count_grid_desc_mblock_mperblock_nblock),
+                            tensor_operation::element_wise::PassThrough,
+                            Sequence<1, PostShuffleThreadSliceSize_M, 1>,
+                            Sequence<0, 1, 2>,
+                            1,
+                            1,
+                            InMemoryDataOperationEnum::Set,
+                            1,
+                            false>{count_grid_desc_mblock_mperblock_nblock,
+                                   mean_var_count_thread_copy_index,
+                                   tensor_operation::element_wise::PassThrough{}};
+
+                        count_thread_copy_vgpr_to_global.Run(
+                            thread_welford_desc_I_m_I,
+                            make_tuple(I0, I0, I0),
+                            count_thread_buf,
+                            count_grid_desc_mblock_mperblock_nblock,
+                            welford_count_grid_buf); // write count
+                    }
+                }
+            });
+
+        } // shuffle C + Ds + welford + write out
+    }     // run
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..69468c25befe036123cf1796798e6a59ea47f74b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp
@@ -0,0 +1,394 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+
+namespace ck {
+
+template <typename EMeanVarDataType,
+          typename HDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename ComputeDataType,
+          typename EHGridDesc_M_N,
+          typename MeanVarGridDesc_M_NBlock,
+          typename CountGridDesc_M_NBlock,
+          typename GammaBetaGridDesc_N,
+          typename HElementwiseOperation,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t NThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t NThreadSliceSize,
+          index_t ESrcVectorSize,
+          index_t HDstVectorSize,
+          index_t GammaSrcVectorSize,
+          index_t BetaSrcVectorSize>
+struct GridwiseWelfordSecondHalfLayernorm2d
+{
+    static_assert(NThreadSliceSize % ESrcVectorSize == 0 &&
+                      NThreadSliceSize % GammaSrcVectorSize == 0 &&
+                      NThreadSliceSize % BetaSrcVectorSize == 0,
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert(NThreadSliceSize % HDstVectorSize == 0,
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    using ThreadClusterLengths_M_N   = Sequence<MThreadClusterSize, NThreadClusterSize>;
+    using ThreadBufferDimAccessOrder = Sequence<0, 1>;
+    using ThreadClusterArrangeOrder  = Sequence<0, 1>;
+
+    static constexpr auto thread_cluster_desc_m_n =
+        make_cluster_descriptor(ThreadClusterLengths_M_N{}, ThreadClusterArrangeOrder{});
+
+    using ThreadBufferLengths_M_N                = Sequence<MThreadSliceSize, NThreadSliceSize>;
+    static constexpr auto thread_buffer_desc_m_n = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<NThreadSliceSize>{}));
+
+    using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
+    static constexpr auto thread_buffer_desc_m_1 =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, Number<1>{}));
+
+    using ThreadBufferLengths_N = Sequence<NThreadSliceSize>;
+    static constexpr auto thread_buffer_desc_n =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<NThreadSliceSize>{}));
+
+    using ThreadWelfordSrcDesc_M_1 = decltype(thread_buffer_desc_m_1);
+    using ThreadWelfordDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using ThreadwiseWelford =
+        ThreadwiseWelfordMerge<ComputeDataType, ThreadWelfordSrcDesc_M_1, ThreadWelfordDstDesc_M>;
+
+    using BlockwiseWelford = BlockwiseWelford<ComputeDataType,
+                                              BlockSize,
+                                              ThreadClusterLengths_M_N,
+                                              ThreadClusterArrangeOrder>;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t N_BlockTileSize = NThreadClusterSize * NThreadSliceSize;
+
+    __device__ static void Run(const EMeanVarDataType* __restrict__ p_e_grid,
+                               const EMeanVarDataType* __restrict__ p_in_welford_mean_grid,
+                               const EMeanVarDataType* __restrict__ p_in_welford_var_grid,
+                               const int32_t* __restrict__ p_in_welford_count_grid,
+                               const GammaDataType* __restrict__ p_gamma_grid,
+                               const BetaDataType* __restrict__ p_beta_grid,
+                               HDataType* __restrict__ p_h_grid,
+                               const EHGridDesc_M_N& e_grid_desc_m_n,
+                               const EHGridDesc_M_N& h_grid_desc_m_n,
+                               const MeanVarGridDesc_M_NBlock& mean_var_grid_desc_m_nblock,
+                               const CountGridDesc_M_NBlock& count_grid_desc_m_nblock,
+                               const GammaBetaGridDesc_N& gamma_grid_desc_n,
+                               const GammaBetaGridDesc_N& beta_grid_desc_n,
+                               index_t numMeanVarCountBlockTileIteration_N,
+                               index_t NBlockClusterLength,
+                               ComputeDataType epsilon,
+                               HElementwiseOperation h_element_op)
+    {
+        // Thread/Block id
+        const index_t thread_local_id = get_thread_local_1d_id();
+        const index_t block_global_id = get_block_1d_id();
+        const auto block_work_idx     = make_tuple(block_global_id / NBlockClusterLength,
+                                               block_global_id % NBlockClusterLength);
+
+        const auto thread_cluster_idx =
+            thread_cluster_desc_m_n.CalculateBottomIndex(make_multi_index(thread_local_id));
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_n_cluster_id = thread_cluster_idx[I1];
+
+        // Global Memory
+        const auto e_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_e_grid, e_grid_desc_m_n.GetElementSpaceSize());
+
+        const auto welford_mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_in_welford_mean_grid, mean_var_grid_desc_m_nblock.GetElementSpaceSize());
+
+        const auto welford_var_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_in_welford_var_grid, mean_var_grid_desc_m_nblock.GetElementSpaceSize());
+
+        const auto welford_count_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_in_welford_count_grid, count_grid_desc_m_nblock.GetElementSpaceSize());
+
+        const auto gamma_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_gamma_grid, gamma_grid_desc_n.GetElementSpaceSize());
+
+        const auto beta_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_beta_grid, beta_grid_desc_n.GetElementSpaceSize());
+
+        auto h_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_h_grid, h_grid_desc_m_n.GetElementSpaceSize());
+
+        // VGPR
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            in_welford_mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            in_welford_var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true>
+            in_welford_count_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            welford_mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            welford_var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true>
+            welford_count_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr,
+                     ComputeDataType,
+                     MThreadSliceSize * NThreadSliceSize,
+                     true>
+            e_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr,
+                     ComputeDataType,
+                     MThreadSliceSize * NThreadSliceSize,
+                     true>
+            gamma_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr,
+                     ComputeDataType,
+                     MThreadSliceSize * NThreadSliceSize,
+                     true>
+            beta_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr,
+                     ComputeDataType,
+                     MThreadSliceSize * NThreadSliceSize,
+                     true>
+            h_thread_buf;
+
+        // IO
+        auto threadwise_mean_load_m_nblock =
+            ThreadwiseTensorSliceTransfer_v2<EMeanVarDataType,
+                                             ComputeDataType,
+                                             MeanVarGridDesc_M_NBlock,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             ThreadBufferDimAccessOrder,
+                                             1,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_grid_desc_m_nblock,
+                make_multi_index(block_work_idx[I0] * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_n_cluster_id));
+
+        auto threadwise_var_load_m_nblock =
+            ThreadwiseTensorSliceTransfer_v2<EMeanVarDataType,
+                                             ComputeDataType,
+                                             MeanVarGridDesc_M_NBlock,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             ThreadBufferDimAccessOrder,
+                                             1,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_grid_desc_m_nblock,
+                make_multi_index(block_work_idx[I0] * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_n_cluster_id));
+
+        auto threadwise_count_load_m_nblock =
+            ThreadwiseTensorSliceTransfer_v2<int32_t,
+                                             int32_t,
+                                             CountGridDesc_M_NBlock,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             ThreadBufferDimAccessOrder,
+                                             1,
+                                             1,
+                                             1,
+                                             true>(
+                count_grid_desc_m_nblock,
+                make_multi_index(block_work_idx[I0] * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_n_cluster_id));
+
+        auto threadwise_e_load_m_n =
+            ThreadwiseTensorSliceTransfer_v2<EMeanVarDataType,
+                                             ComputeDataType,
+                                             decltype(e_grid_desc_m_n),
+                                             decltype(thread_buffer_desc_m_n),
+                                             ThreadBufferLengths_M_N,
+                                             ThreadBufferDimAccessOrder,
+                                             1, // SrcVectorDim
+                                             ESrcVectorSize,
+                                             1,
+                                             true>(
+                e_grid_desc_m_n,
+                make_multi_index(
+                    block_work_idx[I0] * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                    block_work_idx[I1] * N_BlockTileSize + thread_n_cluster_id * NThreadSliceSize));
+
+        auto threadwise_gamma_load_n =
+            ThreadwiseTensorSliceTransfer_v2<GammaDataType,
+                                             ComputeDataType,
+                                             decltype(gamma_grid_desc_n),
+                                             decltype(thread_buffer_desc_n),
+                                             ThreadBufferLengths_N,
+                                             Sequence<0>, // DimAccessOrder,
+                                             0,           // SrcVectorDim,
+                                             GammaSrcVectorSize,
+                                             1,
+                                             true>(
+                gamma_grid_desc_n,
+                make_multi_index(block_work_idx[I1] * N_BlockTileSize +
+                                 thread_n_cluster_id * NThreadSliceSize));
+
+        auto threadwise_beta_load_n =
+            ThreadwiseTensorSliceTransfer_v2<BetaDataType,
+                                             ComputeDataType,
+                                             decltype(beta_grid_desc_n),
+                                             decltype(thread_buffer_desc_n),
+                                             ThreadBufferLengths_N,
+                                             Sequence<0>, // DimAccessOrder,
+                                             0,           // SrcVectorDim,
+                                             BetaSrcVectorSize,
+                                             1,
+                                             true>(
+                beta_grid_desc_n,
+                make_multi_index(block_work_idx[I1] * N_BlockTileSize +
+                                 thread_n_cluster_id * NThreadSliceSize));
+
+        auto threadwise_h_store_m_n =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               HDataType,
+                                               decltype(thread_buffer_desc_m_n),
+                                               decltype(h_grid_desc_m_n),
+                                               HElementwiseOperation,
+                                               ThreadBufferLengths_M_N,
+                                               ThreadBufferDimAccessOrder,
+                                               1, // DstVectorDim
+                                               HDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                h_grid_desc_m_n,
+                make_multi_index(
+                    block_work_idx[I0] * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                    block_work_idx[I1] * N_BlockTileSize + thread_n_cluster_id * NThreadSliceSize),
+                h_element_op);
+
+        // step1: Merge mean and variance
+        constexpr auto mean_var_count_thread_copy_step_I0_n =
+            make_multi_index(I0, NThreadClusterSize);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            welford_mean_thread_buf(I)  = type_convert<ComputeDataType>(0.0f);
+            welford_var_thread_buf(I)   = type_convert<ComputeDataType>(0.0f);
+            welford_count_thread_buf(I) = 0;
+        });
+
+        for(index_t n = 0; n < numMeanVarCountBlockTileIteration_N; ++n)
+        {
+            threadwise_mean_load_m_nblock.Run(mean_var_grid_desc_m_nblock,
+                                              welford_mean_global_val_buf,
+                                              thread_buffer_desc_m_1,
+                                              make_tuple(I0, I0),
+                                              in_welford_mean_thread_buf);
+
+            threadwise_var_load_m_nblock.Run(mean_var_grid_desc_m_nblock,
+                                             welford_var_global_val_buf,
+                                             thread_buffer_desc_m_1,
+                                             make_tuple(I0, I0),
+                                             in_welford_var_thread_buf);
+
+            threadwise_count_load_m_nblock.Run(count_grid_desc_m_nblock,
+                                               welford_count_global_val_buf,
+                                               thread_buffer_desc_m_1,
+                                               make_tuple(I0, I0),
+                                               in_welford_count_thread_buf);
+
+            ThreadwiseWelford::Run(in_welford_mean_thread_buf,
+                                   in_welford_var_thread_buf,
+                                   in_welford_count_thread_buf,
+                                   welford_mean_thread_buf,
+                                   welford_var_thread_buf,
+                                   welford_count_thread_buf);
+
+            threadwise_mean_load_m_nblock.MoveSrcSliceWindow(mean_var_grid_desc_m_nblock,
+                                                             mean_var_count_thread_copy_step_I0_n);
+            threadwise_var_load_m_nblock.MoveSrcSliceWindow(mean_var_grid_desc_m_nblock,
+                                                            mean_var_count_thread_copy_step_I0_n);
+            threadwise_count_load_m_nblock.MoveSrcSliceWindow(count_grid_desc_m_nblock,
+                                                              mean_var_count_thread_copy_step_I0_n);
+        }
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            BlockwiseWelford::Run(
+                welford_mean_thread_buf(I), welford_var_thread_buf(I), welford_count_thread_buf(I));
+        });
+
+        // step2: normalization
+        // h[m, n] = [(e[m, n] - mean[m]) / sqrt(var[m] + eps)] * gamma[n] + beta[n]
+        threadwise_e_load_m_n.Run(e_grid_desc_m_n,
+                                  e_global_val_buf,
+                                  thread_buffer_desc_m_n,
+                                  make_tuple(I0, I0),
+                                  e_thread_buf);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto m) {
+            auto divisor = 1 / ck::math::sqrt(welford_var_thread_buf(m) + epsilon);
+            static_for<0, NThreadSliceSize, 1>{}([&](auto n) {
+                constexpr auto m_n = thread_buffer_desc_m_n.CalculateOffset(make_tuple(m, n));
+                h_thread_buf(Number<m_n>{}) =
+                    (e_thread_buf(Number<m_n>{}) - welford_mean_thread_buf(m)) * divisor;
+            });
+        });
+
+        threadwise_gamma_load_n.Run(gamma_grid_desc_n,
+                                    gamma_global_val_buf,
+                                    thread_buffer_desc_n,
+                                    make_tuple(I0),
+                                    gamma_thread_buf);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto m) {
+            static_for<0, NThreadSliceSize, 1>{}([&](auto n) {
+                constexpr auto m_n = thread_buffer_desc_m_n.CalculateOffset(make_tuple(m, n));
+                h_thread_buf(Number<m_n>{}) = h_thread_buf(Number<m_n>{}) * gamma_thread_buf(n);
+            });
+        });
+
+        threadwise_beta_load_n.Run(beta_grid_desc_n,
+                                   beta_global_val_buf,
+                                   thread_buffer_desc_n,
+                                   make_tuple(I0),
+                                   beta_thread_buf);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto m) {
+            static_for<0, NThreadSliceSize, 1>{}([&](auto n) {
+                constexpr auto m_n = thread_buffer_desc_m_n.CalculateOffset(make_tuple(m, n));
+                h_thread_buf(Number<m_n>{}) = h_thread_buf(Number<m_n>{}) + beta_thread_buf(n);
+            });
+        });
+
+        threadwise_h_store_m_n.Run(thread_buffer_desc_m_n,
+                                   make_tuple(I0, I0),
+                                   h_thread_buf,
+                                   h_grid_desc_m_n,
+                                   h_global_val_buf);
+
+    } // run
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp
index bdebe3816f22d6a4f10e913210abc8c8347e20d0..bd1e0585fc96ad29ee9e5336993f2ba6570fb31d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp
index 1313ec9435e782ad9cc9b92580464f6d5257b93e..fc4f27e33b15b6bfcc7ce942764f9aa5e26cfe1f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp
index 6836a66047531d9eef09dbffc58188f7696a783a..203be3c42d9d23c935d10f34ceb44245f7875ad0 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
index 6c5bd29f9b5ee70b995342f04d59b8ed042b1194..910c926c7e470e0a7daa1cd25f57716f5ce4af6b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,6 +15,7 @@ namespace ck {
 
 template <typename GridwiseReduction,
           bool OutputIndex,
+          bool TransformIndexKtoGlobal,
           bool HaveIndexInput,
           typename InDataType,
           typename OutDataType,
@@ -48,16 +49,17 @@ __global__ void kernel_reduce_threadwise(const InGridDesc_M_K in_grid_desc_m_k,
     }
     else
     {
-        GridwiseReduction::template RunWithIndex<HaveIndexInput>(in_grid_desc_m_k,
-                                                                 out_grid_desc_m,
-                                                                 in_elementwise_op,
-                                                                 acc_elementwise_op,
-                                                                 alpha,
-                                                                 p_in_value_global,
-                                                                 p_in_index_global,
-                                                                 beta,
-                                                                 p_out_value_global,
-                                                                 p_out_index_global);
+        GridwiseReduction::template RunWithIndex<TransformIndexKtoGlobal, HaveIndexInput>(
+            in_grid_desc_m_k,
+            out_grid_desc_m,
+            in_elementwise_op,
+            acc_elementwise_op,
+            alpha,
+            p_in_value_global,
+            p_in_index_global,
+            beta,
+            p_out_value_global,
+            p_out_index_global);
     };
 };
 
@@ -232,7 +234,7 @@ struct GridwiseReduction_mk_to_m_threadwise
             reduced_data_desc, make_tuple(I0), accu_value_buf, out_grid_desc_m, dst_global_buf);
     };
 
-    template <bool HaveIndexInput>
+    template <bool TransformIndexKtoGlobal, bool HaveIndexInput>
     __device__ static void RunWithIndex(const InGridDesc_M_K& in_grid_desc_m_k,
                                         const OutGridDesc_M& out_grid_desc_m,
                                         const InElementwiseOperation& in_elementwise_op,
@@ -390,6 +392,18 @@ struct GridwiseReduction_mk_to_m_threadwise
                 indexStart += KThreadSliceSize;
                 reducedLength += KThreadSliceSize;
             } while(reducedLength < toReduceLength);
+
+            if constexpr(TransformIndexKtoGlobal)
+            {
+                static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                    const auto coord = make_tensor_coordinate(
+                        in_grid_desc_m_k,
+                        make_multi_index(thread_global_1d_id * MThreadSliceSize + I,
+                                         accu_index_buf(I)));
+
+                    accu_index_buf(I) = coord.GetOffset();
+                });
+            }
         };
 
         // for indiced operation, acc_elementwise_op shoud do nothing
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
index fccb127d0f342911ed6bb1f8396c9d9cd0eec921..a8a1f803fcb260e3eba6ccf301f68b7638e2108b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
index b9f4a3080a0e17ca0ba169bc8b83c006acb2305d..59d6bad5d97dacfda3947868aa8c14d0eaf220cc 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..135b9da6a00643c8e74bca6fdf4c36dbcc65f288
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -0,0 +1,1301 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_softmax.hpp"
+
+namespace ck {
+
+template <typename FloatAB,
+          typename FloatGemmAcc,
+          typename FloatCShuffle,
+          typename FloatC,
+          typename D0sDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename C0DEElementwiseOperation,
+          typename B1ElementwiseOperation,
+          typename C1DEElementwiseOperation,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename AGridDesc_AK0_M_AK1,
+          typename BGridDesc_BK0_N_BK1,
+          typename B1GridDesc_BK0_N_BK1,
+          typename C1GridDesc_M_N,
+          typename D0sGridDesc_M_N,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t Gemm1NPerBlock,
+          index_t Gemm1KPerBlock,
+          index_t AK1Value,
+          index_t BK1Value,
+          index_t B1K1Value,
+          index_t MPerXdl,
+          index_t NPerXdl,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          index_t Gemm1NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool AThreadTransferSrcResetCoordinateAfterRun, // ignored
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BThreadTransferSrcResetCoordinateAfterRun, // ignored
+          index_t BBlockLdsExtraN,
+          typename B1BlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename B1BlockTransferThreadClusterArrangeOrder,
+          typename B1BlockTransferSrcAccessOrder,
+          index_t B1BlockTransferSrcVectorDim,
+          index_t B1BlockTransferSrcScalarPerVector,
+          index_t B1BlockTransferDstScalarPerVector_BK1,
+          bool B1ThreadTransferSrcResetCoordinateAfterRun,
+          index_t B1BlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched,
+          bool PadN,
+          bool MaskOutUpperTriangle,
+          int D0sTransferSrcScalarPerVector = 4,
+          PipelineVersion PipelineVer       = PipelineVersion::v1>
+struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
+{
+    static_assert(LoopSched == LoopScheduler::Default,
+                  "Non-default loop scheduler is currently not supported");
+
+    static constexpr index_t NumD0Tensor = D0sDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    // Gemm0
+    static constexpr auto AK0 = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0 = Number<KPerBlock / BK1Value>{};
+    static constexpr auto AK1 = Number<AK1Value>{};
+    static constexpr auto BK1 = Number<BK1Value>{};
+
+    static constexpr auto Gemm0MWaves = MPerBlock / (MPerXdl * MXdlPerWave);
+    static constexpr auto Gemm0NWaves = NPerBlock / (NPerXdl * NXdlPerWave);
+
+    // Gemm1
+    static constexpr auto B1K0 = Number<Gemm1KPerBlock / B1K1Value>{};
+    static constexpr auto B1K1 = Number<B1K1Value>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    using GridwiseGemmPipe = remove_cvref_t<decltype(
+        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+
+    template <typename ABlockDesc_AK0_M_AK1>
+    __host__ __device__ static constexpr auto
+    MakeGemm0AMmaTileDescriptor_M0_M1_M2_K(const ABlockDesc_AK0_M_AK1&)
+    {
+        constexpr index_t MWaves = MPerBlock / (MXdlPerWave * MPerXdl);
+
+        return MakeGemmMmaTileDescriptor_MN0_MN1_MN2_K<MXdlPerWave, MWaves, MPerXdl>(
+            ABlockDesc_AK0_M_AK1{});
+    }
+
+    template <typename BBlockDesc_BK0_N_BK1>
+    __host__ __device__ static constexpr auto
+    MakeGemm0BMmaTileDescriptor_N0_N1_N2_K(const BBlockDesc_BK0_N_BK1&)
+    {
+        constexpr index_t NWaves = NPerBlock / (NXdlPerWave * NPerXdl);
+
+        return MakeGemmMmaTileDescriptor_MN0_MN1_MN2_K<NXdlPerWave, NWaves, NPerXdl>(
+            BBlockDesc_BK0_N_BK1{});
+    }
+
+    template <typename ABlockDesc_AK0_M_AK1>
+    __host__ __device__ static constexpr auto
+    MakeGemm1AMmaTileDescriptor_M0_M1_M2_K(const ABlockDesc_AK0_M_AK1&)
+    {
+        return MakeGemmMmaTileDescriptor_MN0_MN1_MN2_K<MXdlPerWave, 1, 1>(ABlockDesc_AK0_M_AK1{});
+    }
+
+    template <typename BBlockDesc_BK0_N_BK1>
+    __host__ __device__ static constexpr auto
+    MakeGemm1BMmaTileDescriptor_N0_N1_N2_K(const BBlockDesc_BK0_N_BK1&)
+    {
+        constexpr index_t Gemm1NWaves = Gemm1NPerBlock / (Gemm1NXdlPerWave * NPerXdl);
+        return MakeGemmMmaTileDescriptor_MN0_MN1_MN2_K<Gemm1NXdlPerWave, Gemm1NWaves, NPerXdl>(
+            BBlockDesc_BK0_N_BK1{});
+    }
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(AK0, Number<MPerBlock>{}, AK1),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(BK0, Number<NPerBlock>{}, BK1),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetB1BlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B1 matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(B1K0, Number<Gemm1NPerBlock>{}, B1K1),
+            make_tuple(Number<Gemm1NPerBlock + B1BlockLdsExtraN>{} * B1K1, B1K1, I1));
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    {
+        constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+        constexpr index_t NWave = Gemm1NPerBlock / (Gemm1NXdlPerWave * NPerXdl);
+
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl>{},
+                           I1,
+                           Number<CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>{}));
+
+        return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        const index_t gemm0_bytes_end = (SharedMemTrait::a_block_space_size_aligned +
+                                         SharedMemTrait::b_block_space_size_aligned) *
+                                        sizeof(FloatAB);
+        const index_t gemm1_bytes_end =
+            (SharedMemTrait::b1_block_space_offset + SharedMemTrait::b1_block_space_size_aligned) *
+            sizeof(FloatAB);
+        const index_t softmax_bytes_end = (SharedMemTrait::reduction_space_offset +
+                                           SharedMemTrait::reduction_space_size_aligned) *
+                                          sizeof(FloatGemmAcc);
+        const index_t c_block_bytes_end =
+            SharedMemTrait::c_block_space_size * sizeof(FloatCShuffle);
+
+        return math::max(gemm0_bytes_end, gemm1_bytes_end, softmax_bytes_end, c_block_bytes_end);
+    }
+
+    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
+    template <typename Block2CTileMap>
+    __host__ __device__ static constexpr bool
+    CheckValidity(const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
+                  const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
+                  const B1GridDesc_BK0_N_BK1& b1_grid_desc_bk0_n_bk1,
+                  const C1GridDesc_M_N& c1_grid_desc_m_n,
+                  const Block2CTileMap& block_2_ctile_map)
+    {
+        static_assert((MPerBlock % (MPerXdl * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
+                      "Invalid tuning param!");
+
+        const auto M = a_grid_desc_ak0_m_ak1.GetLength(I1);
+        const auto N = b_grid_desc_bk0_n_bk1.GetLength(I1);
+        const auto K = a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2);
+        const auto Gemm1N = b1_grid_desc_bk0_n_bk1.GetLength(I1);
+
+        if(!(M == c1_grid_desc_m_n.GetLength(I0) && Gemm1N == c1_grid_desc_m_n.GetLength(I1)))
+        {
+            return false;
+        }
+
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0 &&
+             Gemm1N % Gemm1NPerBlock == 0))
+        {
+            return false;
+        }
+
+        // check gemm0 gridwise gemm pipeline
+        const auto num_gemm0_k_loop = K / KPerBlock;
+        if(!GridwiseGemmPipe::IsSupported(num_gemm0_k_loop))
+        {
+            return false;
+        }
+
+        // check gemm1 gridwise gemm pipeline
+        if(!(NPerBlock % Gemm1KPerBlock == 0))
+        {
+            return false;
+        }
+
+        const auto num_gemm1_k_inner_loop = NPerBlock / Gemm1KPerBlock;
+        if(!GridwiseGemmPipe::IsSupported(num_gemm1_k_inner_loop))
+        {
+            return false;
+        }
+
+        if(!block_2_ctile_map.CheckValidity(c1_grid_desc_m_n))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / KPerBlock;
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeC1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const C1GridDesc_M_N& c1_grid_desc_m_n)
+    {
+        const auto M = c1_grid_desc_m_n.GetLength(I0);
+        const auto N = c1_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / Gemm1NPerBlock;
+
+        const auto c_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            c1_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<Gemm1NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return c_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // return block_id to C matrix tile idx (m0, n0) mapping
+    __host__ __device__ static constexpr auto
+    MakeDefaultBlock2CTileMap(const C1GridDesc_M_N& c1_grid_desc_m_n)
+    {
+        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, Gemm1NPerBlock, C1GridDesc_M_N>(
+            c1_grid_desc_m_n);
+    }
+
+    __device__ static auto GetGemm0WaveIdx()
+    {
+        const index_t thread_id = get_thread_local_1d_id();
+        constexpr auto WaveSize = MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.wave_size;
+
+        constexpr auto threadid_to_wave_idx_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_merge_transform(make_tuple(Gemm0MWaves, Gemm0NWaves, WaveSize))),
+            make_tuple(Sequence<0, 1, 2>{}),
+            make_tuple(Sequence<0>{}));
+
+        return threadid_to_wave_idx_adaptor.CalculateBottomIndex(make_multi_index(thread_id));
+    }
+
+    __device__ static auto GetGemm0WaveMNIdx(const index_t thread_id)
+    {
+        constexpr auto WaveSize = MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.wave_size;
+        constexpr auto wave_threadid_to_mn_idx_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_merge_transform(make_tuple(WaveSize / MPerXdl, MPerXdl))),
+            make_tuple(Sequence<0, 1>{}),
+            make_tuple(Sequence<0>{}));
+
+        return wave_threadid_to_mn_idx_adaptor.CalculateBottomIndex(make_multi_index(thread_id));
+    }
+
+    static constexpr auto MakeD0sGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
+
+                return static_cast<const D0DataType*>(nullptr);
+            },
+            Number<NumD0Tensor>{});
+    }
+    // D0 desc for source in blockwise copy
+    template <typename D0GridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeGemm0D0GridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(const D0GridDesc_M_N& d0_grid_desc_m_n)
+    {
+        const auto M = d0_grid_desc_m_n.GetLength(I0);
+        const auto N = d0_grid_desc_m_n.GetLength(I1);
+
+        constexpr auto mfma = MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma;
+        constexpr auto N3   = mfma.num_groups_per_blk;
+        constexpr auto N4   = mfma.num_input_blks;
+        constexpr auto N5   = mfma.group_size;
+        return transform_tensor_descriptor(
+            d0_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(
+                           make_tuple(M / MPerBlock, MXdlPerWave, Gemm0MWaves, MPerXdl)),
+                       make_unmerge_transform(
+                           make_tuple(N / NPerBlock, NXdlPerWave, Gemm0NWaves, N3, N4, N5))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 2, 4, 6>{}, Sequence<1, 3, 5, 7, 8, 9>{}));
+    }
+
+    // D0s desc for source in blockwise copy
+    __host__ __device__ static constexpr auto
+    MakeD0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(const D0sGridDesc_M_N& ds_grid_desc_m_n)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeGemm0D0GridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(ds_grid_desc_m_n[i]);
+            },
+            Number<NumD0Tensor>{});
+    }
+
+    using D0sGridPointer                                  = decltype(MakeD0sGridPointer());
+    using D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5 = remove_cvref_t<decltype(
+        MakeD0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(D0sGridDesc_M_N{}))>;
+
+    using C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
+        MakeC1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(C1GridDesc_M_N{}))>;
+
+    using DefaultBlock2CTileMap =
+        remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(C1GridDesc_M_N{}))>;
+
+    struct SharedMemTrait
+    {
+        // LDS allocation for A and B: be careful of alignment
+        static constexpr auto a_block_desc_ak0_m_ak1 =
+            GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        static constexpr auto b_block_desc_bk0_n_bk1 =
+            GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+        static constexpr auto b1_block_desc_bk0_n_bk1 =
+            GetB1BlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        static constexpr auto max_lds_align = math::lcm(math::lcm(AK1, BK1), B1K1);
+
+        static constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+        static constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+        static constexpr auto b1_block_space_size_aligned = math::integer_least_multiple(
+            b1_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+
+        static constexpr auto a_block_space_offset  = 0;
+        static constexpr auto b_block_space_offset  = a_block_space_size_aligned.value;
+        static constexpr auto b1_block_space_offset = 0;
+
+        // LDS allocation for reduction
+        static constexpr index_t reduction_space_size_aligned =
+            math::integer_least_multiple(BlockSize, max_lds_align);
+
+        static constexpr auto reduction_space_offset = 0;
+
+        // LDS allocation for C shuffle in LDS
+        static constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+        static constexpr auto c_block_space_size =
+            c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
+    };
+
+    template <bool HasMainKBlockLoop, typename Block2CTileMap, typename C0MatrixMask>
+    __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
+                               const FloatAB* __restrict__ p_b_grid,
+                               const FloatAB* __restrict__ p_b1_grid,
+                               FloatC* __restrict__ p_c_grid,
+                               D0sGridPointer p_d0s_grid,
+                               void* __restrict__ p_shared,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const C0DEElementwiseOperation& c0de_element_op,
+                               const B1ElementwiseOperation& b1_element_op,
+                               const C1DEElementwiseOperation& c1de_element_op,
+                               const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
+                               const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
+                               const B1GridDesc_BK0_N_BK1& b1_grid_desc_bk0_n_bk1,
+                               const C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   c_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5&
+                                   d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
+                               const Block2CTileMap& block_2_ctile_map,
+                               const C0MatrixMask& c0_matrix_mask)
+    {
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_ak0_m_ak1.GetElementSpaceSize());
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_bk0_n_bk1.GetElementSpaceSize());
+        const auto b1_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b1_grid, b1_grid_desc_bk0_n_bk1.GetElementSpaceSize());
+        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+        const auto d0s_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_d0s_grid[i],
+                    d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i].GetElementSpaceSize());
+            },
+            Number<NumD0Tensor>{});
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        if(!block_2_ctile_map.ValidCTileIndex(
+               block_work_idx,
+               make_tuple(c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
+                          c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
+        {
+            return;
+        }
+
+        // HACK: this force m/gemm1_n_block_data_idx_on_grid into SGPR
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+
+        const index_t gemm1_n_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * Gemm1NPerBlock);
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        //
+        // set up Gemm0
+        //
+
+        // A matrix blockwise copy
+        auto a_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                AElementwiseOperation,
+                                                tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<AK0, MPerBlock, AK1>,
+                                                ABlockTransferThreadClusterLengths_AK0_M_AK1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                FloatAB,
+                                                FloatAB,
+                                                decltype(a_grid_desc_ak0_m_ak1),
+                                                decltype(a_block_desc_ak0_m_ak1),
+                                                ABlockTransferSrcAccessOrder,
+                                                Sequence<1, 0, 2>,
+                                                ABlockTransferSrcVectorDim,
+                                                2,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_AK1,
+                                                1,
+                                                1,
+                                                true, // SrcResetCoord
+                                                true, // DstResetCoord
+                                                NumGemmKPrefetchStage>(
+                a_grid_desc_ak0_m_ak1,
+                make_multi_index(0, m_block_data_idx_on_grid, 0),
+                a_element_op,
+                a_block_desc_ak0_m_ak1,
+                make_multi_index(0, 0, 0),
+                tensor_operation::element_wise::PassThrough{});
+
+        // B matrix blockwise copy
+        auto b_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<BK0, NPerBlock, BK1>,
+                                                BBlockTransferThreadClusterLengths_BK0_N_BK1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                FloatAB,
+                                                FloatAB,
+                                                decltype(b_grid_desc_bk0_n_bk1),
+                                                decltype(b_block_desc_bk0_n_bk1),
+                                                BBlockTransferSrcAccessOrder,
+                                                Sequence<1, 0, 2>,
+                                                BBlockTransferSrcVectorDim,
+                                                2,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_BK1,
+                                                1,
+                                                1,
+                                                true, // SrcResetCoord
+                                                true, // DstResetCoord
+                                                NumGemmKPrefetchStage>(
+                b_grid_desc_bk0_n_bk1,
+                make_multi_index(0, 0, 0), // will loop over GemmN dimension
+                b_element_op,
+                b_block_desc_bk0_n_bk1,
+                make_multi_index(0, 0, 0),
+                tensor_operation::element_wise::PassThrough{});
+
+        // Fused Gemm+Gemm pipeline
+        // for n in N0:
+        //   for k in K0:
+        //     acc[m][n] += A[m][k] * B0[k][n]
+        //   acc1[m][o] += acc[m][n] * B1[n][o]
+
+        // sanity check
+        constexpr index_t KPack = math::max(
+            math::lcm(AK1, BK1), MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+
+        auto blockwise_gemm = BlockwiseGemmXdlops_v2<
+            BlockSize,
+            FloatAB,
+            FloatGemmAcc,
+            decltype(a_block_desc_ak0_m_ak1),
+            decltype(b_block_desc_bk0_n_bk1),
+            decltype(MakeGemm0AMmaTileDescriptor_M0_M1_M2_K(a_block_desc_ak0_m_ak1)),
+            decltype(MakeGemm0BMmaTileDescriptor_N0_N1_N2_K(b_block_desc_bk0_n_bk1)),
+            MPerBlock,
+            NPerBlock,
+            KPerBlock,
+            MPerXdl,
+            NPerXdl,
+            MXdlPerWave,
+            NXdlPerWave,
+            KPack,
+            true>{}; // TransposeC
+
+        auto acc_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+        // LDS allocation for A and B: be careful of alignment
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<FloatAB*>(p_shared) + SharedMemTrait::a_block_space_offset,
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<FloatAB*>(p_shared) + SharedMemTrait::b_block_space_offset,
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1, 0, 0);
+        const auto a_block_reset_copy_step =
+            make_multi_index(-a_grid_desc_ak0_m_ak1.GetLength(I0), 0, 0);
+        const auto b_block_reset_copy_step =
+            make_multi_index(-b_grid_desc_bk0_n_bk1.GetLength(I0), NPerBlock, 0);
+
+        // gridwise GEMM pipeline
+        // Only supports LoopScheduler::Default
+        const auto gridwise_gemm_pipeline = GridwiseGemmPipeline_Selector<PipelineVer,
+                                                                          NumGemmKPrefetchStage,
+                                                                          LoopScheduler::Default>();
+
+        const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
+            (a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2)) /
+            KPerBlock);
+
+        //
+        // set up Gemm1
+        //
+
+        // Acc matrix threadwise copy: AccVGPR to VGPR and downcast to XDL input data type
+        constexpr auto acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4 =
+            blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4();
+
+        constexpr auto m0 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I0);
+        constexpr auto n0 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I1);
+        constexpr auto m1 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I2);
+        constexpr auto n1 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I3);
+        constexpr auto m2 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I4);
+        constexpr auto n2 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I5);
+        constexpr auto n3 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I6);
+        constexpr auto n4 = acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I7);
+
+        constexpr auto b1_block_slice_copy_step = make_multi_index(Gemm1KPerBlock / B1K1, 0, 0);
+
+        // d0 matrix threadwise copy
+        constexpr auto d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5 =
+            make_naive_tensor_descriptor_packed(make_tuple(I1,   // MBlockId
+                                                           I1,   // NBlockID
+                                                           m0,   // MRepeat
+                                                           n0,   // NRepeat
+                                                           m1,   // MWaveId
+                                                           n1,   // NWaveId
+                                                           m2,   // MPerXdl
+                                                           n2,   // NGroupNum
+                                                           n3,   // NInputNum
+                                                           n4)); // registerNum
+
+        auto d0s_thread_buf = generate_tuple(
+            [&](auto i) {
+                using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
+                return StaticBuffer<
+                    AddressSpaceEnum::Vgpr,
+                    D0DataType,
+                    d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5.GetElementSpaceSize(),
+                    true>{};
+            },
+            Number<NumD0Tensor>{});
+
+        const auto wave_id     = GetGemm0WaveIdx();
+        const auto wave_m_n_id = GetGemm0WaveMNIdx(wave_id[I2]); // I2: 0~63
+
+        auto d0s_threadwise_copy = generate_tuple(
+            [&](auto i) {
+                using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
+                return ThreadwiseTensorSliceTransfer_v2<
+                    D0DataType,
+                    D0DataType,
+                    decltype(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i]),
+                    decltype(d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5),
+                    Sequence<I1, // MBlockId
+                             I1, // NBlockID
+                             m0, // MRepeat
+                             n0, // NRepeat
+                             m1, // MWaveId
+                             n1, // NWaveId
+                             m2, // MPerXdl
+                             n2, // NGroupNum
+                             n3, // NInputNum
+                             n4>,
+                    Sequence<0, 1, 2, 3, 4, 5, 6, 7, 8, 9>,
+                    9,
+                    D0sTransferSrcScalarPerVector,
+                    1,
+                    false>(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
+                           make_multi_index(block_work_idx[I0], // MBlockId
+                                            0,                  // NBlockId
+                                            0,                  // mrepeat
+                                            0,                  // nrepeat
+                                            wave_id[I0],        // MWaveId
+                                            wave_id[I1],        // NWaveId
+                                            wave_m_n_id[I1],    // MPerXdl
+                                            0,                  // group
+                                            wave_m_n_id[I0],    // NInputIndex
+                                            0));                // register number
+            },
+            Number<NumD0Tensor>{});
+        // acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4 to acc_thread_desc_k0_m_k1
+        // n0_n1_n2_n3 -> k0
+        // m0_m1_m2 -> m
+        // n4 -> k1
+        // NOTE: had to use merge_v3 or will spit out compilation errors
+        constexpr auto acc_thread_desc_k0_m_k1 = transform_tensor_descriptor(
+            acc_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4,
+            make_tuple(make_merge_transform_v3_division_mod(make_tuple(n0, n1, n2, n3)),
+                       make_merge_transform_v3_division_mod(make_tuple(m0, m1, m2)),
+                       make_pass_through_transform(n4)),
+            make_tuple(Sequence<1, 3, 5, 6>{}, Sequence<0, 2, 4>{}, Sequence<7>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+        // A1 matrix in AccVGPR
+        // N2 num_groups_per_blk, N3 num_input_blks, N4 group_size
+        constexpr auto AccN3 =
+            blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_N2_N3_N4().GetLength(I6);
+
+        constexpr auto A1ThreadSlice_K0_M_K1 =
+            make_tuple(Number<Gemm1KPerBlock / n4 / AccN3>{}, Number<m0 * m1 * m2>{}, Number<n4>{});
+
+        constexpr auto A1ThreadSliceK0        = A1ThreadSlice_K0_M_K1[I0];
+        constexpr auto A1ThreadSliceM         = A1ThreadSlice_K0_M_K1[I1];
+        constexpr auto A1ThreadSliceK1        = A1ThreadSlice_K0_M_K1[I2];
+        constexpr auto a1_thread_desc_k0_m_k1 = make_naive_tensor_descriptor(
+            A1ThreadSlice_K0_M_K1,
+            make_tuple(A1ThreadSliceM * A1ThreadSliceK1, A1ThreadSliceK1, I1));
+
+        // B1 matrix in LDS memory, dst of blockwise copy
+        constexpr auto b1_block_desc_bk0_n_bk1 = GetB1BlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // A1 matrix blockwise copy
+        auto a1_blockwise_copy = ThreadwiseTensorSliceTransfer_StaticToStatic<
+            FloatGemmAcc,
+            FloatAB,
+            decltype(acc_thread_desc_k0_m_k1),
+            decltype(a1_thread_desc_k0_m_k1),
+            tensor_operation::element_wise::PassThrough,
+            Sequence<A1ThreadSliceK0, A1ThreadSliceM, A1ThreadSliceK1>,
+            Sequence<1, 0, 2>,
+            2,
+            n4>{tensor_operation::element_wise::PassThrough{}};
+
+        // B1 matrix blockwise copy
+        auto b1_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<B1K0, Gemm1NPerBlock, B1K1>,
+                                                B1BlockTransferThreadClusterLengths_BK0_N_BK1,
+                                                B1BlockTransferThreadClusterArrangeOrder,
+                                                FloatAB,
+                                                FloatAB,
+                                                decltype(b1_grid_desc_bk0_n_bk1),
+                                                decltype(b1_block_desc_bk0_n_bk1),
+                                                B1BlockTransferSrcAccessOrder,
+                                                Sequence<1, 0, 2>,
+                                                B1BlockTransferSrcVectorDim,
+                                                2,
+                                                B1BlockTransferSrcScalarPerVector,
+                                                B1BlockTransferDstScalarPerVector_BK1,
+                                                1,
+                                                1,
+                                                B1ThreadTransferSrcResetCoordinateAfterRun,
+                                                true, // DstResetCoord
+                                                NumGemmKPrefetchStage>(
+                b1_grid_desc_bk0_n_bk1,
+                make_multi_index(0, gemm1_n_block_data_idx_on_grid, 0),
+                b1_element_op,
+                b1_block_desc_bk0_n_bk1,
+                make_multi_index(0, 0, 0),
+                tensor_operation::element_wise::PassThrough{});
+
+        auto a1_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAB>(
+            a1_thread_desc_k0_m_k1.GetElementSpaceSize());
+
+        // reuse LDS space for gemm0's b_block_buf
+        auto b1_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<FloatAB*>(p_shared) + SharedMemTrait::b1_block_space_offset,
+            b1_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        // selected_mfma.group_size or B1K1 <= Gemm1KPack <= selected_mfma.group_size
+        // selected_mfma.k_per_blk <= Gemm1KPack
+        //
+        // Following similar rationale behind Gemm0KPack, let Gemm1KPack be the lowest common
+        // multiples of A1K1 (predetermined by selected_mfma.group_size) and B1K1. But in this case
+        // Gemm1KPack can't be higher than A1K1 itself because A1 matrix is distributed in VGPRs
+        // with 'group_size' amount of contiguous elements. Having Gemm1KPack greater than A1K1 will
+        // cause mismatch in summation index for example c[0:7] = a1[[0:3, 8:11]] * b1[0:7].
+        // therefore we may just as well assign Gemm1KPack = group_size
+        constexpr index_t Gemm1KPack =
+            MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.group_size;
+
+        auto gemm1_blockwise_gemm = BlockwiseGemmXdlops_v2<
+            BlockSize,
+            FloatAB,
+            FloatGemmAcc,
+            decltype(a1_thread_desc_k0_m_k1),
+            decltype(b1_block_desc_bk0_n_bk1),
+            decltype(MakeGemm1AMmaTileDescriptor_M0_M1_M2_K(a1_thread_desc_k0_m_k1)),
+            decltype(MakeGemm1BMmaTileDescriptor_N0_N1_N2_K(b1_block_desc_bk0_n_bk1)),
+            MPerBlock,
+            Gemm1NPerBlock,
+            Gemm1KPerBlock,
+            MPerXdl,
+            NPerXdl,
+            MXdlPerWave,
+            Gemm1NXdlPerWave,
+            Gemm1KPack,
+            true,       // TransposeC
+            Gemm1KPack, // AMmaKStride
+            Gemm1KPack * XdlopsGemm<FloatAB, MPerXdl, NPerXdl, Gemm1KPack, false>{}.K0PerXdlops>{
+            // BMmaKStride
+            make_tuple(0, 0, 0, 0)}; // A_origin
+
+        auto acc1_thread_buf = gemm1_blockwise_gemm.GetCThreadBuffer();
+
+        //
+        // Blockwise softmax
+        //
+        auto workspace_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<FloatGemmAcc*>(p_shared) + SharedMemTrait::reduction_space_offset,
+            SharedMemTrait::reduction_space_size_aligned);
+
+        // get acc0 8D thread cluster
+        constexpr auto thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4 =
+            blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_N2_N3_N4().GetLengths() /
+            blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4().GetLengths();
+        constexpr auto tm0 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I0);
+        constexpr auto tn0 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I1);
+        constexpr auto tm1 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I2);
+        constexpr auto tn1 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I3);
+        constexpr auto tm2 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I4);
+        constexpr auto tn2 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I5);
+        constexpr auto tn3 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I6);
+        constexpr auto tn4 = thread_cluster_m0_n0_m1_n1_m2_n2_n3_n4.At(I7);
+
+        // get acc0 thread map
+        constexpr auto m0_n_m1_to_m_n_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_unmerge_transform(make_tuple(tm0 * tm1, tm2)),
+                       make_pass_through_transform(I1)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        constexpr auto threadid_to_m0_n_m1_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(
+                make_merge_transform(make_tuple(tm0 * tm1, tn0 * tn1 * tn2 * tn3 * tn4, tm2))),
+            make_tuple(Sequence<0, 1, 2>{}),
+            make_tuple(Sequence<0>{}));
+        const auto threadid_to_m_n_thread_cluster_adaptor =
+            chain_tensor_adaptors(m0_n_m1_to_m_n_adaptor, threadid_to_m0_n_m1_adaptor);
+
+        // get acc0 2D thread cluster & 2D thread slice
+        constexpr auto thread_cluster_desc_m_n = make_naive_tensor_descriptor_packed(
+            make_tuple(tm0 * tm1 * tm2, tn0 * tn1 * tn2 * tn3 * tn4));
+        constexpr auto thread_slice_desc_m_n =
+            make_naive_tensor_descriptor_packed(make_tuple(m0 * m1 * m2, n0 * n1 * n2 * n3 * n4));
+
+        auto blockwise_softmax = BlockwiseSoftmax<BlockSize,
+                                                  FloatGemmAcc,
+                                                  decltype(threadid_to_m_n_thread_cluster_adaptor),
+                                                  decltype(thread_cluster_desc_m_n),
+                                                  decltype(thread_slice_desc_m_n)>{};
+
+        const index_t num_gemm1_k_block_outer_loop =
+            b_grid_desc_bk0_n_bk1.GetLength(I1) / NPerBlock;
+        constexpr index_t num_gemm1_k_block_inner_loop = NPerBlock / Gemm1KPerBlock;
+
+        // Initialize C
+        StaticBuffer<AddressSpaceEnum::Vgpr, FloatGemmAcc, acc1_thread_buf.Size(), true>
+            c_thread_buf;
+        c_thread_buf.Clear();
+
+        // Initialize running sum and max of exponentiating row vectors
+        using SoftmaxBuf = typename decltype(blockwise_softmax)::BufferType;
+        SoftmaxBuf running_sum, running_sum_new, running_max, running_max_new;
+        running_sum     = 0;
+        running_sum_new = 0;
+        running_max     = NumericLimits<FloatGemmAcc>::Lowest();
+        running_max_new = NumericLimits<FloatGemmAcc>::Lowest();
+
+        // gemm1 K loop
+        index_t gemm1_k_block_outer_index = 0;
+        do
+        {
+            auto n_block_data_idx_on_grid =
+                __builtin_amdgcn_readfirstlane(gemm1_k_block_outer_index * NPerBlock);
+            if(c0_matrix_mask.IsTileSkippable(
+                   m_block_data_idx_on_grid, n_block_data_idx_on_grid, MPerBlock, NPerBlock))
+            {
+                continue;
+            }
+            // gemm0
+            gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_grid_desc_ak0_m_ak1,
+                                                                   a_block_desc_ak0_m_ak1,
+                                                                   a_blockwise_copy,
+                                                                   a_grid_buf,
+                                                                   a_block_buf,
+                                                                   a_block_slice_copy_step,
+                                                                   b_grid_desc_bk0_n_bk1,
+                                                                   b_block_desc_bk0_n_bk1,
+                                                                   b_blockwise_copy,
+                                                                   b_grid_buf,
+                                                                   b_block_buf,
+                                                                   b_block_slice_copy_step,
+                                                                   blockwise_gemm,
+                                                                   acc_thread_buf,
+                                                                   num_k_block_main_loop);
+            // multiple d
+            if constexpr(NumD0Tensor)
+            {
+                static_assert(NXdlPerWave == n0);
+                static_assert(MXdlPerWave == m0);
+
+                static_for<0, NumD0Tensor, 1>{}([&](auto i) {
+                    d0s_threadwise_copy(i).Run(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
+                                               d0s_grid_buf[i],
+                                               d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
+                                               make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
+                                               d0s_thread_buf(i));
+                });
+                static_for<0, m0 * n0 * n2 * n4, 1>{}([&](auto i) {
+                    // get reference to src data
+                    const auto src_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iSrc) -> const auto& { return d0s_thread_buf[iSrc][i]; },
+                        Number<NumD0Tensor>{});
+
+                    // get reference to dst data
+                    auto dst_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto) -> auto& { return acc_thread_buf(i); },
+                        Number<2>{});
+
+                    unpack2(c0de_element_op, dst_data_refs, src_data_refs);
+                });
+                static_for<0, NumD0Tensor, 1>{}([&](auto i) {
+                    d0s_threadwise_copy(i).MoveSrcSliceWindow(
+                        d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
+                        make_multi_index(0, 1, 0, 0, 0, 0, 0, 0, 0, 0));
+                });
+            }
+            else
+            {
+                static_for<0, acc_thread_buf.Size(), 1>{}(
+                    [&](auto i) { c0de_element_op(acc_thread_buf(i), acc_thread_buf[i]); });
+            }
+
+            // do MNK padding or upper triangular masking
+            if constexpr(MaskOutUpperTriangle || PadN)
+            {
+                // 8d thread_desc in thread scope
+                constexpr auto c_thread_lengths =
+                    blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4().GetLengths();
+
+                // 8d block_desc in block scope
+                constexpr auto c_block_lengths =
+                    blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_N2_N3_N4().GetLengths();
+
+                constexpr auto M0 = c_block_lengths[I0];
+                constexpr auto N0 = c_block_lengths[I1];
+                constexpr auto M1 = c_block_lengths[I2];
+                constexpr auto N1 = c_block_lengths[I3];
+                constexpr auto M2 = c_block_lengths[I4];
+                constexpr auto N2 = c_block_lengths[I5];
+                constexpr auto N3 = c_block_lengths[I6];
+                constexpr auto N4 = c_block_lengths[I7];
+
+                // works like multi-dimension static_for (static_ford), but provides both the linear
+                // index as well as n-d index
+                using Acc0TileIterator = SpaceFillingCurve<
+                    decltype(c_thread_lengths),
+                    typename arithmetic_sequence_gen<0, c_thread_lengths.Size(), 1>::type,
+                    typename uniform_sequence_gen<c_thread_lengths.Size(), 1>::type,
+                    false>; // SnakeCurved
+
+                auto acc0_thread_origin = blockwise_gemm.CalculateCThreadOriginDataIndex8D(
+                    Number<0>{}, Number<0>{}, Number<0>{}, Number<0>{});
+
+                constexpr auto block_idx_to_m_n_adaptor = make_single_stage_tensor_adaptor(
+                    make_tuple(make_unmerge_transform(make_tuple(M0, M1, M2)),
+                               make_unmerge_transform(make_tuple(N0, N1, N2, N3, N4))),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0, 2, 4>{}, Sequence<1, 3, 5, 6, 7>{}));
+
+                static_for<0, Acc0TileIterator::GetNumOfAccess(), 1>{}([&](auto i) {
+                    auto acc0_thread_idx = Acc0TileIterator::GetIndex(i) + acc0_thread_origin;
+                    auto m_local =
+                        block_idx_to_m_n_adaptor.CalculateBottomIndex(acc0_thread_idx)[I0];
+                    auto n_local =
+                        block_idx_to_m_n_adaptor.CalculateBottomIndex(acc0_thread_idx)[I1];
+                    auto m_global = m_local + m_block_data_idx_on_grid;
+                    auto n_global = n_local + n_block_data_idx_on_grid;
+                    if(c0_matrix_mask.IsMaskedElement(m_global, n_global))
+                    {
+                        acc_thread_buf(i) = -ck::NumericLimits<float>::Infinity();
+                    }
+                });
+            }
+
+            block_sync_lds(); // wait for lds read in gemm0 blockwise gemm
+
+            // softmax
+            SoftmaxBuf& max = blockwise_softmax.max_value_buf;
+            SoftmaxBuf& sum = blockwise_softmax.sum_value_buf;
+
+            blockwise_softmax.Run(acc_thread_buf, workspace_buf);
+
+            // TODO: may convert to log domain
+            running_max_new = mathext::max(max, running_max);
+            running_sum_new = mathext::exp(running_max - running_max_new) * running_sum +
+                              mathext::exp(max - running_max_new) * sum;
+
+            // gemm1
+            {
+                // TODO: explore using dynamic buffer for a1 thread buffer
+                // For a1_blockwise_copy, the goal is to satisfy pipeline requirements RunRead(),
+                // RunWrite(), and MoveSliceWindow(). But it is impossible to implement given that
+                // the A1 source buffer is static buffer holding the output of first GEMM and
+                // requires constexpr offset by design. Therefore, we pass tensor coordinate offset
+                // explicitly in Run() below.
+
+                // Initialize acc1
+                acc1_thread_buf.Clear();
+
+                // preload data into LDS
+                b1_blockwise_copy.RunRead(b1_grid_desc_bk0_n_bk1, b1_grid_buf);
+
+                b1_blockwise_copy.MoveSrcSliceWindow(b1_grid_desc_bk0_n_bk1,
+                                                     b1_block_slice_copy_step);
+
+                block_sync_lds(); // wait for reduction LDS read
+
+                b1_blockwise_copy.RunWrite(b1_block_desc_bk0_n_bk1, b1_block_buf);
+
+                // main body
+                if constexpr(num_gemm1_k_block_inner_loop > 1)
+                {
+                    static_for<0, num_gemm1_k_block_inner_loop - 1, 1>{}([&](auto i) {
+                        a1_blockwise_copy.Run(acc_thread_desc_k0_m_k1,
+                                              make_tuple(Number<i * A1ThreadSliceK0>{}, I0, I0),
+                                              acc_thread_buf,
+                                              a1_thread_desc_k0_m_k1,
+                                              make_tuple(I0, I0, I0),
+                                              a1_thread_buf);
+
+                        b1_blockwise_copy.RunRead(b1_grid_desc_bk0_n_bk1, b1_grid_buf);
+
+                        block_sync_lds();
+
+                        gemm1_blockwise_gemm.Run(a1_thread_buf, b1_block_buf, acc1_thread_buf);
+
+                        block_sync_lds();
+
+                        b1_blockwise_copy.MoveSrcSliceWindow(b1_grid_desc_bk0_n_bk1,
+                                                             b1_block_slice_copy_step);
+
+                        b1_blockwise_copy.RunWrite(b1_block_desc_bk0_n_bk1, b1_block_buf);
+                    });
+                }
+                // tail
+                {
+                    a1_blockwise_copy.Run(
+                        acc_thread_desc_k0_m_k1,
+                        make_tuple(
+                            Number<(num_gemm1_k_block_inner_loop - 1) * A1ThreadSliceK0>{}, I0, I0),
+                        acc_thread_buf,
+                        a1_thread_desc_k0_m_k1,
+                        make_tuple(I0, I0, I0),
+                        a1_thread_buf);
+
+                    block_sync_lds();
+
+                    gemm1_blockwise_gemm.Run(a1_thread_buf, b1_block_buf, acc1_thread_buf);
+                }
+            } // end gemm1
+
+            constexpr auto c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4 =
+                gemm1_blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4();
+            constexpr auto cm0 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I0);
+            constexpr auto cn0 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I1);
+            constexpr auto cm1 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I2);
+            constexpr auto cn1 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I3);
+            constexpr auto cm2 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I4);
+            constexpr auto cn2 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I5);
+            constexpr auto cn3 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I6);
+            constexpr auto cn4 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I7);
+            constexpr auto c_thread_slice_desc_m_n = make_naive_tensor_descriptor_packed(
+                make_tuple(cm0 * cm1 * cm2, cn0 * cn1 * cn2 * cn3 * cn4));
+            constexpr auto c_thread_buf_slice_m = c_thread_slice_desc_m_n.GetLength(I0);
+            constexpr auto c_thread_buf_slice_n = c_thread_slice_desc_m_n.GetLength(I1);
+
+            static_for<0, c_thread_buf_slice_m, 1>{}([&](auto iM) {
+                static_for<0, c_thread_buf_slice_n, 1>{}([&](auto iN) {
+                    auto I = Number<c_thread_slice_desc_m_n.CalculateOffset(make_tuple(iM, iN))>{};
+                    FloatGemmAcc acc1 = acc1_thread_buf[I]; // P*V
+                    FloatGemmAcc c    = c_thread_buf[I];    // O
+                    FloatGemmAcc c_new =
+                        (running_sum[iM] * math::exp(running_max[iM] - running_max_new[iM]) * c +
+                         math::exp(max[iM] - running_max_new[iM]) * acc1) /
+                        running_sum_new[iM]; // Formula by Dao et al.,
+                                             // https://arxiv.org/pdf/2205.14135v2.pdf section 3.1
+
+                    c_thread_buf(I) = c_new; // O_new
+                });
+            });
+
+            a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_ak0_m_ak1,
+                                                a_block_reset_copy_step); // rewind K
+            b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_bk0_n_bk1,
+                                                b_block_reset_copy_step); // rewind K and step N
+
+            // update before next j iteration
+            running_max = running_max_new;
+            running_sum = running_sum_new;
+
+            block_sync_lds(); // wait for gemm1 LDS read
+        } while(++gemm1_k_block_outer_index < num_gemm1_k_block_outer_loop); // end j loop
+
+        // shuffle C and write out
+        {
+            static_assert(MXdlPerWave % CShuffleMXdlPerWavePerShuffle == 0 &&
+                              Gemm1NXdlPerWave % CShuffleNXdlPerWavePerShuffle == 0,
+                          "wrong!");
+
+            constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+            constexpr index_t NWave = Gemm1NPerBlock / (Gemm1NXdlPerWave * NPerXdl);
+
+            // TODO: hacky, fix it!
+            constexpr auto c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4 =
+                gemm1_blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4();
+
+            // TODO: hacky, fix it!
+            // c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp is only used to get lengths
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp =
+                gemm1_blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_N2_N3_N4();
+
+            constexpr auto M0 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I0);
+            constexpr auto N0 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I1);
+            constexpr auto M1 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I2);
+            constexpr auto N1 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I3);
+            constexpr auto M2 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I4);
+            constexpr auto N2 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I5);
+            constexpr auto N3 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I6);
+            constexpr auto N4 = c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4_tmp.GetLength(I7);
+
+            constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+                GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+            auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                static_cast<FloatCShuffle*>(p_shared),
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4 = transform_tensor_descriptor(
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleMXdlPerWavePerShuffle>{}, // M0 (MXdlPerWave) per shuffle
+                        M1,                                      // M1 = MWave
+                        M2)),                                    // M2 = MPerXdl
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleNXdlPerWavePerShuffle>{}, // N0 (NXdlPerWave) per shuffle
+                        N1,                                      // N1 = NWave
+                        N2,                                      // N2 * N3 * N4 = NPerXdl
+                        N3,
+                        N4))),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(
+                    Sequence<>{}, Sequence<0, 2, 4>{}, Sequence<>{}, Sequence<1, 3, 5, 6, 7>{}));
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block =
+                gemm1_blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
+
+            const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+            const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_block_to_m0_m1_m2_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(M0, M1, M2))),
+                    make_tuple(Sequence<0, 1, 2>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto m_thread_data_on_block_idx =
+                m_thread_data_on_block_to_m0_m1_m2_adaptor.CalculateBottomIndex(
+                    make_multi_index(m_thread_data_on_block));
+
+            const auto n_thread_data_on_block_to_n0_n1_n2_n3_n4_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(N0, N1, N2, N3, N4))),
+                    make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_block_idx =
+                n_thread_data_on_block_to_n0_n1_n2_n3_n4_adaptor.CalculateBottomIndex(
+                    make_multi_index(n_thread_data_on_block));
+
+            // shuffle: threadwise copy C from VGPR to LDS
+            auto c_thread_copy_vgpr_to_lds =
+                ThreadwiseTensorSliceTransfer_v1r3<FloatGemmAcc,
+                                                   FloatCShuffle,
+                                                   decltype(c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4),
+                                                   decltype(c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4),
+                                                   tensor_operation::element_wise::PassThrough,
+                                                   Sequence<CShuffleMXdlPerWavePerShuffle,
+                                                            CShuffleNXdlPerWavePerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            I1,
+                                                            N2,
+                                                            I1,
+                                                            N4>,
+                                                   Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                                   7,
+                                                   1,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>{
+                    c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4,
+                    make_multi_index(0,
+                                     0,
+                                     m_thread_data_on_block_idx[I1],
+                                     n_thread_data_on_block_idx[I1],
+                                     m_thread_data_on_block_idx[I2],
+                                     n_thread_data_on_block_idx[I2],
+                                     n_thread_data_on_block_idx[I3],
+                                     n_thread_data_on_block_idx[I4]),
+                    tensor_operation::element_wise::PassThrough{}};
+
+            // shuffle: blockwise copy C from LDS to global
+            auto c_shuffle_block_copy_lds_to_global = ThreadGroupTensorSliceTransfer_v6r1<
+                ThisThreadBlock,            // ThreadGroup
+                C1DEElementwiseOperation,   // ElementwiseOperation,
+                CGlobalMemoryDataOperation, // DstInMemOp,
+                Sequence<1,
+                         CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                         1,
+                         CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>, // BlockSliceLengths,
+                CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                FloatCShuffle,        // typename SrcData,
+                FloatC,               // typename DstData,
+                decltype(c_shuffle_block_desc_mblock_mperblock_nblock_nperblock),
+                decltype(c_grid_desc_mblock_mperblock_nblock_nperblock),
+                Sequence<0, 1, 2, 3>,                           // typename DimAccessOrder,
+                3,                                              // index_t VectorDim,
+                CShuffleBlockTransferScalarPerVector_NPerBlock, // index_t ScalarPerVector,
+                true,  // bool ThreadTransferSrcResetCoordinateAfterRun,
+                false> // bool ThreadTransferDstResetCoordinateAfterRun>
+                {c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                 make_multi_index(0, 0, 0, 0),
+                 c_grid_desc_mblock_mperblock_nblock_nperblock,
+                 make_multi_index(block_work_idx[I0], 0, block_work_idx[I1], 0),
+                 c1de_element_op};
+
+            // space filling curve for threadwise C in VGPR
+            constexpr auto sfc_c_vgpr =
+                SpaceFillingCurve<Sequence<MXdlPerWave, Gemm1NXdlPerWave, 1, 1, 1, N2, 1, N4>,
+                                  Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                  Sequence<CShuffleMXdlPerWavePerShuffle,
+                                           CShuffleNXdlPerWavePerShuffle,
+                                           1,
+                                           1,
+                                           1,
+                                           N2,
+                                           1,
+                                           N4>>{};
+
+            // space filling curve for shuffled blockwise C in global mem
+            constexpr auto sfc_c_global =
+                SpaceFillingCurve<Sequence<1, MPerBlock, 1, Gemm1NPerBlock>,
+                                  Sequence<0, 2, 1, 3>,
+                                  Sequence<1,
+                                           CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                                           1,
+                                           CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>>{};
+
+            constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+            static_assert(num_access == sfc_c_global.GetNumOfAccess(), "wrong!");
+
+            static_for<0, num_access, 1>{}([&](auto access_id) {
+                // make sure it's safe to write to LDS
+                block_sync_lds();
+
+                // each thread write its data from VGPR to LDS
+                c_thread_copy_vgpr_to_lds.Run(c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4,
+                                              sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                              c_thread_buf,
+                                              c_block_desc_m0_n0_m1_n1_m2_n2_n3_n4,
+                                              c_shuffle_block_buf);
+
+                // make sure it's safe to read from LDS
+                block_sync_lds();
+
+                // each block copy its data from LDS to global
+                c_shuffle_block_copy_lds_to_global.Run(
+                    c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                    c_shuffle_block_buf,
+                    c_grid_desc_mblock_mperblock_nblock_nperblock,
+                    c_grid_buf);
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    constexpr auto c_global_step = sfc_c_global.GetForwardStep(access_id);
+
+                    // move on C
+                    c_shuffle_block_copy_lds_to_global.MoveDstSliceWindow(
+                        c_grid_desc_mblock_mperblock_nblock_nperblock, c_global_step);
+                }
+            });
+        }
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
index fec360b7fa0ee916648b8eefc67ef3da0590d042..9eb2bf8aa81ea2689a4a8c71c2b78b070e701a4f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -18,6 +18,10 @@
 
 namespace ck {
 
+/**
+ * @brief Gridwise gemm + softmax + gemm fusion
+ *
+ */
 template <typename FloatAB,
           typename FloatGemmAcc,
           typename FloatCShuffle,
@@ -879,14 +883,6 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
                 }
             } // end gemm1
 
-            // workaround compiler issue; see ck/ck.hpp
-            if constexpr(CK_WORKAROUND_SWDEV_XXXXXX_BF16_ATTEN_FWD_GFX908_ISSUE == 1 &&
-                         is_same_v<FloatAB, bhalf_t> && MPerBlock == 256 && NPerBlock == 128 &&
-                         Gemm1NPerBlock == 128)
-            {
-                __builtin_amdgcn_sched_barrier(0);
-            }
-
             constexpr auto c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4 =
                 gemm1_blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4();
             constexpr auto cm0 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I0);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v1.hpp
index 8dfd3490f24e65b76b7a520f71ea6849a71c2aaf..78fa011306775936c5ff63c14063bbafc813100d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v1.hpp
@@ -20,6 +20,10 @@
 
 namespace ck {
 
+/**
+ * @brief Gridwise gemm + softmax + gemm fusion
+ *
+ */
 template <typename FloatAB,
           typename ZDataType,
           typename FloatGemm,
@@ -1119,14 +1123,6 @@ struct GridwiseBatchedMultiheadAttentionForward_Xdl_CShuffle_V1
                 }
             } // end gemm1
 
-            // workaround compiler issue; see ck/ck.hpp
-            if constexpr(CK_WORKAROUND_SWDEV_XXXXXX_BF16_ATTEN_FWD_GFX908_ISSUE == 1 &&
-                         (is_same_v<FloatGemm, bhalf_t>)&&MPerBlock == 256 && NPerBlock == 128 &&
-                         Gemm1NPerBlock == 128)
-            {
-                __builtin_amdgcn_sched_barrier(0);
-            }
-
             constexpr auto c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4 =
                 gemm1_blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4();
             constexpr auto cm0 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I0);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v2.hpp
index 0b3eb018e83b9e3935c38ace0e396853a547dd65..8e381a0ad8a3e589a1f8207502f4f88e56a699d7 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_mha_fwd_xdl_cshuffle_v2.hpp
@@ -20,6 +20,10 @@
 
 namespace ck {
 
+/**
+ * @brief Gridwise gemm + softmax + gemm fusion
+ *
+ */
 template <typename FloatAB,
           typename ZDataType,
           typename FloatGemm,
@@ -1297,14 +1301,6 @@ struct GridwiseBatchedMultiheadAttentionForward_Xdl_CShuffle_V2
                 }
             } // end gemm1
 
-            // workaround compiler issue; see ck/ck.hpp
-            if constexpr(CK_WORKAROUND_SWDEV_XXXXXX_BF16_ATTEN_FWD_GFX908_ISSUE == 1 &&
-                         (is_same_v<FloatGemm, bhalf_t>)&&MPerBlock == 256 && NPerBlock == 128 &&
-                         Gemm1NPerBlock == 128)
-            {
-                __builtin_amdgcn_sched_barrier(0);
-            }
-
             constexpr auto c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4 =
                 gemm1_blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2_N3_N4();
             constexpr auto cm0 = c_thread_desc_m0_n0_m1_n1_m2_n2_n3_n4.GetLength(I0);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp
index ede6a96dc9f53045b5fc029f4c4f4f6aa4fa4817..ed1ffdd85765e84feb2f6824a858ab842487c18c 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp
index 33c45a0f037b4c81fbaf810cc10e55f71a2ab254..b6c83af13a18639cf7f8312dd416dbf4d9cb5297 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_contraction_dlops_v1r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_contraction_dlops_v1r2.hpp
deleted file mode 100644
index 2369f51795d9a3bcf5a28ad9de702d75e24fac48..0000000000000000000000000000000000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_contraction_dlops_v1r2.hpp
+++ /dev/null
@@ -1,662 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_CONTRACTION_DLOPS_V1R2_HPP
-#define CK_GRIDWISE_CONTRACTION_DLOPS_V1R2_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_gemm_dlops_v2r3.hpp"
-#include "blockwise_tensor_slice_transfer_v2.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_set.hpp"
-
-namespace ck {
-
-template <typename GridwiseContraction,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_GK0_GM0_GM10_GM11_GK1,
-          typename BGridDesc_GK0_GN0_GN10_GN11_GK1,
-          typename CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1,
-          typename CGridBlockCluster_BlockId_To_GM10_GN10,
-          bool HasMainKBlockLoop,
-          bool HasDoubleTailKBlockLoop>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_contraction_dlops_v1r2(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AGridDesc_GK0_GM0_GM10_GM11_GK1 a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-            const BGridDesc_GK0_GN0_GN10_GN11_GK1 b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-            const CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1 c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-            const CGridBlockCluster_BlockId_To_GM10_GN10 c_grid_block_cluster_blockid_to_gm10_gn10)
-{
-    constexpr index_t shared_block_size =
-        GridwiseContraction::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseContraction::Run(p_a_grid,
-                             p_b_grid,
-                             p_c_grid,
-                             p_shared_block,
-                             a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                             b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                             c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                             c_grid_block_cluster_blockid_to_gm10_gn10,
-                             integral_constant<bool, HasMainKBlockLoop>{},
-                             integral_constant<bool, HasDoubleTailKBlockLoop>{});
-}
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_GK0_GM0_GM1_GK1,
-          typename BGridDesc_GK0_GN0_GN1_GK1,
-          typename CGridDesc_GM0_GM1_GN0_GN1,
-          index_t GM1PerBlockGM11,
-          index_t GN1PerBlockGN11,
-          index_t GK0PerBlock,
-          index_t BM1PerThreadBM11,
-          index_t BN1PerThreadBN11,
-          index_t BK0PerThread,
-          typename BM10BN10ThreadClusterBM10Xs,
-          typename BM10BN10ThreadClusterBN10Xs,
-          typename ABlockTransferThreadSliceLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferThreadClusterLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          typename ABlockTransferSrcVectorTensorLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferDstVectorTensorLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
-          typename BBlockTransferThreadSliceLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferThreadClusterLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferThreadClusterArrangeOrder,
-          typename BBlockTransferSrcAccessOrder,
-          typename BBlockTransferSrcVectorTensorLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferDstVectorTensorLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGridStepHacks,
-          typename BGridStepHacks,
-          typename CGridStepHacks,
-          typename AGridMoveSliceWindowStepHacks,
-          typename BGridMoveSliceWindowStepHacks>
-struct GridwiseContractionDlops_A_GK0_GM0_GM1_GK1_B_GK0_GN0_GN1_GK1_C_GM0_GM1_GN0_GN1
-{
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-
-    // GM0 and GN0 need to known at compile-time
-    static constexpr auto GM0 = CGridDesc_GM0_GM1_GN0_GN1{}.GetLength(I0);
-    static constexpr auto GN0 = CGridDesc_GM0_GM1_GN0_GN1{}.GetLength(I2);
-    static constexpr auto GK1 = AGridDesc_GK0_GM0_GM1_GK1{}.GetLength(I3);
-
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        // lds max alignment
-        // TODO: part of them should be moved into blockwise-gemm
-        // TODO: change this. I think it needs multi-dimensional alignment
-        constexpr auto max_lds_align = GK1;
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_block_desc_gk0_gm0_gm10_gm11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GM0, I1, Number<GM1PerBlockGM11>{}, GK1),
-            max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_block_desc_gk0_gn0_gn10_gn11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GN0, I1, Number<GN1PerBlockGN11>{}, GK1),
-            max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size = math::integer_least_multiple(
-            a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size = math::integer_least_multiple(
-            b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        return 2 * (a_block_aligned_space_size + b_block_aligned_space_size) * sizeof(FloatAB);
-    }
-
-    __host__ __device__ static constexpr bool
-    CheckValidity(const AGridDesc_GK0_GM0_GM1_GK1& a_grid_desc_gk0_gm0_gm1_gk1,
-                  const BGridDesc_GK0_GN0_GN1_GK1& b_grid_desc_gk0_gn0_gn1_gk1,
-                  const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        static_assert(is_known_at_compile_time<remove_cv_t<decltype(GM0)>>::value &&
-                          is_known_at_compile_time<remove_cv_t<decltype(GN0)>>::value,
-                      "wrong! GM0 and GN0 need to be known at compile-time");
-
-        const auto GM1 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I2);
-        const auto GN1 = b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I2);
-        const auto GK0 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I0);
-
-        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
-
-        return (
-            (GM0 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I0) &&
-             GM1 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1) &&
-             GN0 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I2) &&
-             GN1 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3) &&
-             GM0 == a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I1) &&
-             GM1 == a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I2) &&
-             GN0 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I1) &&
-             GN1 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I2) &&
-             GK0 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I0) &&
-             GK1 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I3)) &&
-            (GM1 % GM1PerBlockGM11 == 0 && GN1 % GN1PerBlockGN11 == 0 && GK0 % GK0PerBlock == 0));
-    }
-
-    __host__ __device__ static constexpr index_t
-    CalculateGridSize(const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        const auto GM1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1);
-        const auto GN1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3);
-
-        constexpr index_t GM11 = GM1PerBlockGM11;
-        constexpr index_t GN11 = GN1PerBlockGN11;
-
-        const index_t GM10 = GM1 / GM11;
-        const index_t GN10 = GN1 / GN11;
-
-        const index_t grid_size = GM10 * GN10;
-
-        return grid_size;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t GK0)
-    {
-        const bool has_main_k_block_loop = (GK0 + GK0PerBlock) / (2 * GK0PerBlock) > 1;
-
-        return has_main_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasDoubleTailKBlockLoop(index_t GK0)
-    {
-        const bool has_double_tail_k_block_loop = (GK0 / GK0PerBlock) % 2 == 0;
-
-        return has_double_tail_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr auto MakeAGridDescriptor_GK0_GM0_GM10_GM11_GK1(
-        const AGridDesc_GK0_GM0_GM1_GK1& a_grid_desc_gk0_gm0_gm1_gk1)
-    {
-        const auto GK0 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I0);
-        const auto GM1 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I2);
-
-        const auto GM11 = Number<GM1PerBlockGM11>{};
-        const auto GM10 = GM1 / GM11;
-
-        const auto a_grid_desc_gk0_gm0_gm10_gm11_gk1 = transform_tensor_descriptor(
-            a_grid_desc_gk0_gm0_gm1_gk1,
-            make_tuple(make_pass_through_transform(GK0),
-                       make_pass_through_transform(GM0),
-                       make_unmerge_transform(make_tuple(GM10, GM11)),
-                       make_pass_through_transform(GK1)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-        return a_grid_desc_gk0_gm0_gm10_gm11_gk1;
-    }
-
-    __host__ __device__ static constexpr auto MakeBGridDescriptor_GK0_GN0_GN10_GN11_GK1(
-        const BGridDesc_GK0_GN0_GN1_GK1& b_grid_desc_gk0_gn0_gn1_gk1)
-    {
-        const auto GK0 = b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I0);
-        const auto GN1 = b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I2);
-
-        const auto GN11 = Number<GN1PerBlockGN11>{};
-        const auto GN10 = GN1 / GN11;
-
-        const auto b_grid_desc_gk0_gn0_gn10_gn11_gk1 = transform_tensor_descriptor(
-            b_grid_desc_gk0_gn0_gn1_gk1,
-            make_tuple(make_pass_through_transform(GK0),
-                       make_pass_through_transform(GN0),
-                       make_unmerge_transform(make_tuple(GN10, GN11)),
-                       make_pass_through_transform(GK1)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-        return b_grid_desc_gk0_gn0_gn10_gn11_gk1;
-    }
-
-    __host__ __device__ static constexpr auto MakeCGridDescriptor_GM10_BM0_BM1_GN10_BN0_BN1(
-        const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        const auto GM1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1);
-        const auto GN1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3);
-
-        constexpr auto GM11 = Number<GM1PerBlockGM11>{};
-        constexpr auto GN11 = Number<GN1PerBlockGN11>{};
-
-        const auto GM10 = GM1 / GM11;
-        const auto GN10 = GN1 / GN11;
-
-        constexpr auto BM = GM0 * GM11;
-        constexpr auto BN = GN0 * GN11;
-
-        constexpr auto BM1 =
-            Number<container_reduce(BM10BN10ThreadClusterBM10Xs{}, math::multiplies{}, I1) *
-                   BM1PerThreadBM11>{};
-        constexpr auto BN1 =
-            Number<container_reduce(BM10BN10ThreadClusterBN10Xs{}, math::multiplies{}, I1) *
-                   BN1PerThreadBN11>{};
-
-        constexpr auto BM0 = BM / BM1;
-        constexpr auto BN0 = BN / BN1;
-
-        const auto c_gm0_gm10_gm11_gn0_gn10_gn11_grid_desc = transform_tensor_descriptor(
-            c_grid_desc_gm0_gm1_gn0_gn1,
-            make_tuple(make_pass_through_transform(GM0),
-                       make_unmerge_transform(make_tuple(GM10, GM11)),
-                       make_pass_through_transform(GN0),
-                       make_unmerge_transform(make_tuple(GN10, GN11))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}, Sequence<4, 5>{}));
-
-        const auto c_gm10_bm_gn10_bn_grid_desc = transform_tensor_descriptor(
-            c_gm0_gm10_gm11_gn0_gn10_gn11_grid_desc,
-            make_tuple(make_pass_through_transform(GM10),
-                       make_merge_transform(make_tuple(GM0, GM11)),
-                       make_pass_through_transform(GN10),
-                       make_merge_transform(make_tuple(GN0, GN11))),
-            make_tuple(Sequence<1>{}, Sequence<0, 2>{}, Sequence<4>{}, Sequence<3, 5>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-        const auto c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1 = transform_tensor_descriptor(
-            c_gm10_bm_gn10_bn_grid_desc,
-            make_tuple(make_pass_through_transform(GM10),
-                       make_unmerge_transform(make_tuple(BM0, BM1)),
-                       make_pass_through_transform(GN10),
-                       make_unmerge_transform(make_tuple(BN0, BN1))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}, Sequence<4, 5>{}));
-
-        return c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1;
-    }
-
-    __host__ __device__ static constexpr auto MakeCGridBlockCluster_BlockId_To_GM10_GN10(
-        const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        const auto GM1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1);
-        const auto GN1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3);
-
-        constexpr auto GM11 = Number<GM1PerBlockGM11>{};
-        constexpr auto GN11 = Number<GN1PerBlockGN11>{};
-
-        const auto GM10 = GM1 / GM11;
-        const auto GN10 = GN1 / GN11;
-
-        const auto c_grid_block_cluster_blockid_to_gm10_gn10 = make_single_stage_tensor_adaptor(
-            make_tuple(make_merge_transform(make_tuple(GM10, GN10))),
-            make_tuple(Sequence<0, 1>{}),
-            make_tuple(Sequence<0>{}));
-
-        return c_grid_block_cluster_blockid_to_gm10_gn10;
-    }
-
-    using AGridDesc_GK0_GM0_GM10_GM11_GK1 =
-        decltype(MakeAGridDescriptor_GK0_GM0_GM10_GM11_GK1(AGridDesc_GK0_GM0_GM1_GK1{}));
-    using BGridDesc_GK0_GN0_GN10_GN11_GK1 =
-        decltype(MakeBGridDescriptor_GK0_GN0_GN10_GN11_GK1(BGridDesc_GK0_GN0_GN1_GK1{}));
-    using CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1 =
-        decltype(MakeCGridDescriptor_GM10_BM0_BM1_GN10_BN0_BN1(CGridDesc_GM0_GM1_GN0_GN1{}));
-    using CGridBlockCluster_BlockId_To_GM10_GN10 =
-        decltype(MakeCGridBlockCluster_BlockId_To_GM10_GN10(CGridDesc_GM0_GM1_GN0_GN1{}));
-
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ static void
-    Run(const FloatAB* __restrict__ p_a_grid,
-        const FloatAB* __restrict__ p_b_grid,
-        FloatC* __restrict__ p_c_grid,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_GK0_GM0_GM10_GM11_GK1& a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-        const BGridDesc_GK0_GN0_GN10_GN11_GK1& b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-        const CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1& c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-        const CGridBlockCluster_BlockId_To_GM10_GN10& c_grid_block_cluster_blockid_to_gm10_gn10,
-        integral_constant<bool, HasMainKBlockLoop>,
-        integral_constant<bool, HasDoubleTailKBlockLoop>)
-    {
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_grid, a_grid_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_grid, b_grid_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize());
-        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_grid, c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1.GetElementSpaceSize());
-
-        const auto GK0 = a_grid_desc_gk0_gm0_gm10_gm11_gk1.GetLength(I0);
-
-        // divide block work by [GM10, GN10]
-        const auto c_gm10_gn10_block_cluster_idx =
-            c_grid_block_cluster_blockid_to_gm10_gn10.CalculateBottomIndex(
-                make_multi_index(get_block_1d_id()));
-
-        // HACK: this force index data into SGPR
-        const index_t igm10 = __builtin_amdgcn_readfirstlane(c_gm10_gn10_block_cluster_idx[I0]);
-        const index_t ign10 = __builtin_amdgcn_readfirstlane(c_gm10_gn10_block_cluster_idx[I1]);
-
-        // lds max alignment
-        // TODO: part of them should be moved into blockwise-gemm
-        // TODO: change this. I think it needs multi-dimensional alignment
-        constexpr auto max_lds_align = GK1;
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_block_desc_gk0_gm0_gm10_gm11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GM0, I1, Number<GM1PerBlockGM11>{}, GK1),
-            max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_block_desc_gk0_gn0_gn10_gn11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GN0, I1, Number<GN1PerBlockGN11>{}, GK1),
-            max_lds_align);
-
-        // A matrix in LDS memory for blockwise GEMM
-        //   be careful of LDS alignment
-        constexpr auto a_block_desc_gk0_bm_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GM0 * Number<GM1PerBlockGM11>{}, GK1), max_lds_align);
-
-        // B matrix in LDS memory for blockwise GEMM
-        //   be careful of LDS alignment
-        constexpr auto b_block_desc_gk0_bn_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GN0 * Number<GN1PerBlockGN11>{}, GK1), max_lds_align);
-
-        static_assert(a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize() ==
-                              a_block_desc_gk0_bm_gk1.GetElementSpaceSize() &&
-                          b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize() ==
-                              b_block_desc_gk0_bn_gk1.GetElementSpaceSize(),
-                      "wrong!");
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy = BlockwiseTensorSliceTransfer_v5r1<
-            BlockSize,
-            InMemoryDataOperationEnum::Set,
-            Sequence<GK0PerBlock, GM0, 1, GM1PerBlockGM11, GK1.value>,
-            ABlockTransferThreadSliceLengths_GK0_GM0_GM10_GM11_GK1,
-            ABlockTransferThreadClusterLengths_GK0_GM0_GM10_GM11_GK1,
-            ABlockTransferThreadClusterArrangeOrder,
-            FloatAB,
-            FloatAB,
-            decltype(a_grid_desc_gk0_gm0_gm10_gm11_gk1),
-            decltype(a_block_desc_gk0_gm0_gm10_gm11_gk1),
-            ABlockTransferSrcAccessOrder,
-            Sequence<0, 1, 2, 3, 4>,
-            ABlockTransferSrcVectorTensorLengths_GK0_GM0_GM10_GM11_GK1, // SrcVectorTensorLengths
-            ABlockTransferDstVectorTensorLengths_GK0_GM0_GM10_GM11_GK1, // DstVectorTensorLengths
-            ABlockTransferSrcVectorTensorContiguousDimOrder, // SrcVectorTensorContiguousDimOrder
-            Sequence<0, 1, 2, 3, 4>,                         // DstVectorTensorContiguousDimOrder
-            false,
-            true>(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                  make_multi_index(0, 0, igm10, 0, 0),
-                  a_block_desc_gk0_gm0_gm10_gm11_gk1,
-                  make_multi_index(0, 0, 0, 0, 0));
-
-        // B matrix blockwise copy
-        auto b_blockwise_copy = BlockwiseTensorSliceTransfer_v5r1<
-            BlockSize,
-            InMemoryDataOperationEnum::Set,
-            Sequence<GK0PerBlock, GN0, 1, GN1PerBlockGN11, GK1.value>,
-            BBlockTransferThreadSliceLengths_GK0_GN0_GN10_GN11_GK1,
-            BBlockTransferThreadClusterLengths_GK0_GN0_GN10_GN11_GK1,
-            BBlockTransferThreadClusterArrangeOrder,
-            FloatAB,
-            FloatAB,
-            decltype(b_grid_desc_gk0_gn0_gn10_gn11_gk1),
-            decltype(b_block_desc_gk0_gn0_gn10_gn11_gk1),
-            BBlockTransferSrcAccessOrder,
-            Sequence<0, 1, 2, 3, 4>,
-            BBlockTransferSrcVectorTensorLengths_GK0_GN0_GN10_GN11_GK1, // SrcVectorTensorLengths
-            BBlockTransferDstVectorTensorLengths_GK0_GN0_GN10_GN11_GK1, // DstVectorTensorLengths
-            BBlockTransferSrcVectorTensorContiguousDimOrder, // SrcVectorTensorContiguousDimOrder
-            Sequence<0, 1, 2, 3, 4>,                         // DstVectorTensorContiguousDimOrder
-            false,
-            true>(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                  make_multi_index(0, 0, ign10, 0, 0),
-                  b_block_desc_gk0_gn0_gn10_gn11_gk1,
-                  make_multi_index(0, 0, 0, 0, 0));
-
-        // GEMM definition
-        //   c_mtx += transpose(a_mtx) * b_mtx
-        //     a_mtx[GK0PerBlock, GM1PerBlockGM11] is in LDS
-        //     b_mtx[KPerBlocl, GN1PerBlockGN11] is in LDS
-        //     c_mtx[GM1PerBlockGM11, GN1PerBlockGN11] is distributed among threads, and saved in
-        //       register
-        const auto blockwise_gemm =
-            BlockwiseGemmDlops_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_pipeline_BM0_2_BN0_2<
-                BlockSize,
-                FloatAB,
-                FloatAB,
-                FloatAcc,
-                decltype(a_block_desc_gk0_bm_gk1),
-                decltype(b_block_desc_gk0_bn_gk1),
-                BM1PerThreadBM11,
-                BN1PerThreadBN11,
-                BK0PerThread,
-                BM10BN10ThreadClusterBM10Xs,
-                BM10BN10ThreadClusterBN10Xs,
-                BM1PerThreadBM11,
-                BN1PerThreadBN11>{};
-
-        constexpr auto c_thread_tensor_lengths_bm0_bm1_bn0_bn1 =
-            decltype(blockwise_gemm)::GetCThreadTensorLengths_BM0_BM1_BN0_BN1();
-
-        constexpr auto c_thread_desc_bm0_bm1_bn0_bn1 = make_naive_tensor_descriptor_packed(
-            sequence_to_tuple_of_number(c_thread_tensor_lengths_bm0_bm1_bn0_bn1));
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size = math::integer_least_multiple(
-            a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size = math::integer_least_multiple(
-            b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        FloatAB* p_a_block_double = p_shared_block;
-        FloatAB* p_b_block_double = p_shared_block + 2 * a_block_aligned_space_size;
-
-        // register allocation for output
-        auto c_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAcc>(
-            c_thread_desc_bm0_bm1_bn0_bn1.GetElementSpaceSize());
-
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_thread_desc_bm0_bm1_bn0_bn1),
-                                    decltype(c_thread_tensor_lengths_bm0_bm1_bn0_bn1)>{}
-            .Run(c_thread_desc_bm0_bm1_bn0_bn1,
-                 make_tuple(I0, I0, I0, I0),
-                 c_thread_buf,
-                 FloatAcc{0});
-
-        constexpr auto a_block_slice_copy_step = make_multi_index(GK0PerBlock, 0, 0, 0, 0);
-        constexpr auto b_block_slice_copy_step = make_multi_index(GK0PerBlock, 0, 0, 0, 0);
-
-        auto a_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double, a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize());
-        auto b_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double, b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize());
-
-        auto a_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double + a_block_aligned_space_size,
-            a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize());
-        auto b_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double + b_block_aligned_space_size,
-            b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize());
-
-        // LDS double buffer: preload data into LDS
-        {
-            a_blockwise_copy.RunRead(
-                a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-            b_blockwise_copy.RunRead(
-                b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-            a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_even_buf);
-            b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_even_buf);
-        }
-
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t gk0_block_on_grid = 0;
-
-            // LDS double buffer: main body
-            // use Do-While loop instead of For loop to simplify control flow
-            do
-            {
-                // even iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                                                    a_block_slice_copy_step,
-                                                    AGridMoveSliceWindowStepHacks{});
-                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                                                    b_block_slice_copy_step,
-                                                    BGridMoveSliceWindowStepHacks{});
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-                b_blockwise_copy.RunRead(
-                    b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(c_thread_desc_bm0_bm1_bn0_bn1,
-                                   a_block_even_buf,
-                                   b_block_even_buf,
-                                   c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_odd_buf);
-                b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_odd_buf);
-
-                // odd iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                                                    a_block_slice_copy_step,
-                                                    AGridMoveSliceWindowStepHacks{});
-                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                                                    b_block_slice_copy_step,
-                                                    BGridMoveSliceWindowStepHacks{});
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-                b_blockwise_copy.RunRead(
-                    b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(
-                    c_thread_desc_bm0_bm1_bn0_bn1, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_even_buf);
-                b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_even_buf);
-
-                gk0_block_on_grid += 2 * GK0PerBlock;
-            } while(gk0_block_on_grid < GK0 - 2 * GK0PerBlock);
-        }
-
-        // LDS double buffer: tail
-        if constexpr(HasDoubleTailKBlockLoop) // if has 2 iteration left
-        {
-            a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                                                a_block_slice_copy_step,
-                                                AGridMoveSliceWindowStepHacks{});
-            b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                                                b_block_slice_copy_step,
-                                                BGridMoveSliceWindowStepHacks{});
-
-            __syncthreads();
-
-            // LDS double buffer: load last data from device mem
-            a_blockwise_copy.RunRead(
-                a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-            b_blockwise_copy.RunRead(
-                b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-            // LDS double buffer: GEMM on 2nd-last data
-            blockwise_gemm.Run(
-                c_thread_desc_bm0_bm1_bn0_bn1, a_block_even_buf, b_block_even_buf, c_thread_buf);
-
-            // LDS double buffer: store last data to LDS
-            a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_odd_buf);
-            b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_odd_buf);
-
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_thread_desc_bm0_bm1_bn0_bn1, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-        }
-        else // if has 1 iteration left
-        {
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_thread_desc_bm0_bm1_bn0_bn1, a_block_even_buf, b_block_even_buf, c_thread_buf);
-        }
-
-        // output: register to global memory
-        {
-            constexpr auto c_thread_desc_gm10_bm0_bm1_gn10_bn0_bn1 =
-                make_naive_tensor_descriptor_packed(
-                    make_tuple(I1,
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I0]>{},
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I1]>{},
-                               I1,
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I2]>{},
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I3]>{}));
-
-            const auto c_thread_origin_on_block_bm0_bm1_bn0_bn1 =
-                blockwise_gemm.CalculateCThreadOriginOnBlock_BM0_BM1_BN0_BN1(
-                    get_thread_local_1d_id());
-
-            ThreadwiseTensorSliceTransfer_v1r3<
-                FloatAcc,
-                FloatC,
-                decltype(c_thread_desc_gm10_bm0_bm1_gn10_bn0_bn1),
-                decltype(c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1),
-                Sequence<1,
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I0],
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I1],
-                         1,
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I2],
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I3]>,
-                CThreadTransferSrcDstAccessOrder,
-                CThreadTransferSrcDstVectorDim,
-                CThreadTransferDstScalarPerVector,
-                CGlobalMemoryDataOperation,
-                1,
-                false>{c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                       make_multi_index(igm10,
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I0],
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I1],
-                                        ign10,
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I2],
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I3])}
-                .Run(c_thread_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                     make_tuple(I0, I0, I0, I0, I0, I0),
-                     c_thread_buf,
-                     c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                     c_grid_buf,
-                     CGridStepHacks{});
-        }
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp
index 8b82b65540d1c5612002549f72708be07b649dae..d686c14b350953a48bf132694251ac677fb47c4f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp
index 05257d16275e5f9817ec5d9cbbd06b133f0b5e05..bf0e8c186c75cef05b08c7ef9e1bf16c016830a3 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 //
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp
index 0d5cbca925af85eb661fe852e238aa8b09e5e717..3ea72b85345869a938e52f872c2c586c6e36170f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -434,7 +434,7 @@ struct GridwiseElementwiseLayernormWelfordVariance_mk_to_mk
             });
 
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                auto divisor = 1 / __builtin_amdgcn_sqrtf(var_thread_buf(iM) + epsilon);
+                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                 static_for<0, XThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
index 16ba23280ddcdf9c3f1d87cb2e4cd64a72691650..533559adf173813a239b522cc5214889ceae07b9 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -66,7 +66,8 @@ __global__ void
             const ReduceGridDescriptor_MBlock_MPerBlock reduce_grid_desc_mblock_mperblock,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp
index a9522a66917525385d8fe58c7450daca5e534543..27f48a84ba72fe284a964ac295540eac1c4f6766 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -185,8 +185,10 @@ struct GridwiseGemmDlMultipleD_km_kn_mn
         return b_grid_desc_k0_n0_n1_k1;
     }
 
+    // E desc for destination in blockwise copy
+    template <typename CGridDesc_M_N_>
     __host__ __device__ static constexpr auto
-    MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(const CGridDesc_M_N& c_grid_desc_m_n)
+    MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(const CGridDesc_M_N_& c_grid_desc_m_n)
     {
         const auto M = c_grid_desc_m_n.GetLength(I0);
         const auto N = c_grid_desc_m_n.GetLength(I1);
@@ -238,19 +240,19 @@ struct GridwiseGemmDlMultipleD_km_kn_mn
     using BGridDesc_K0_N0_N1_K1 = decltype(MakeBGridDescriptor_K0_N0_N1_K1(BGridDesc_K0_N_K1{}));
     using CGridDesc_M0_M10_M11_N0_N10_N11 =
         decltype(MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(CGridDesc_M_N{}));
-    using Block2CTileMap = decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}));
 
     using DsGridPointer = decltype(MakeDsGridPointer());
 
     template <typename DsGridDesc_M0_M10_M11_N0_N10_N11,
               bool HasMainKBlockLoop,
-              bool HasDoubleTailKBlockLoop>
+              bool HasDoubleTailKBlockLoop,
+              typename Block2CTileMap>
     __device__ static void
     Run(const FloatAB* __restrict__ p_a_grid,
         const FloatAB* __restrict__ p_b_grid,
         DsGridPointer p_ds_grid,
         FloatC* __restrict__ p_c_grid,
-        FloatAB* __restrict__ p_shared_block,
+        void* __restrict__ p_shared_block,
         const AElementwiseOperation&,
         const BElementwiseOperation&,
         const CDEElementwiseOperation& cde_element_op,
@@ -399,8 +401,9 @@ struct GridwiseGemmDlMultipleD_km_kn_mn
         constexpr auto b_block_aligned_space_size = math::integer_least_multiple(
             b_block_desc_k0_n0_n1_k1.GetElementSpaceSize(), max_lds_align);
 
-        FloatAB* p_a_block_double = p_shared_block;
-        FloatAB* p_b_block_double = p_shared_block + 2 * a_block_aligned_space_size;
+        FloatAB* p_a_block_double = static_cast<FloatAB*>(p_shared_block);
+        FloatAB* p_b_block_double =
+            static_cast<FloatAB*>(p_shared_block) + 2 * a_block_aligned_space_size;
 
         // register allocation for output
         auto c_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAcc>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
index c839cde0f70aa7c23cf7fe80e688294cbc11a445..d46aea5e22d131a588c90fc36189c34bd5755242 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -574,4 +574,546 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
     }
 };
 
+template <index_t BlockSize,
+          typename FloatAB,
+          typename FloatAcc,
+          typename FloatC,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename AGridDesc_B_K0_M_K1,
+          typename BGridDesc_B_K0_N_K1,
+          typename CGridDesc_M_N,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t K1Value,
+          index_t M1PerThreadM111,
+          index_t N1PerThreadN111,
+          index_t KPerThread,
+          typename M11N11ThreadClusterM110Xs,
+          typename M11N11ThreadClusterN110Xs,
+          typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
+          typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+          typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
+          typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector>
+struct GridwiseGemmDl_bkm_bkn_mn_v1r3
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+
+    // K1 should be Number<...>
+    static constexpr auto K1 = Number<K1Value>{};
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // TODO: change this. I think it needs multi-dimensional alignment
+        constexpr auto max_lds_align = K1;
+
+        // TODO: check alignment
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_b_k0_m_k1 = make_naive_tensor_descriptor_aligned(
+            make_tuple(Number<1>{}, Number<K0PerBlock>{}, Number<MPerBlock>{}, K1), max_lds_align);
+
+        // TODO: check alignment
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_b_k0_n_k1 = make_naive_tensor_descriptor_aligned(
+            make_tuple(Number<1>{}, Number<K0PerBlock>{}, Number<NPerBlock>{}, K1), max_lds_align);
+
+        // TODO: check alignment
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_aligned_space_size = math::integer_least_multiple(
+            a_block_desc_b_k0_m_k1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_aligned_space_size = math::integer_least_multiple(
+            b_block_desc_b_k0_n_k1.GetElementSpaceSize(), max_lds_align);
+
+        return 2 * (a_block_aligned_space_size + b_block_aligned_space_size) * sizeof(FloatAB);
+    }
+
+    __host__ __device__ static constexpr bool
+    CheckValidity(const AGridDesc_B_K0_M_K1& a_grid_desc_b_k0_m_k1,
+                  const BGridDesc_B_K0_N_K1& b_grid_desc_b_k0_n_k1,
+                  const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        const auto M      = a_grid_desc_b_k0_m_k1.GetLength(I2);
+        const auto N      = b_grid_desc_b_k0_n_k1.GetLength(I2);
+        const auto K0     = a_grid_desc_b_k0_m_k1.GetLength(I1);
+        const auto KBatch = a_grid_desc_b_k0_m_k1.GetLength(I0);
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+
+        return (M == c_grid_desc_m_n.GetLength(I0) && N == c_grid_desc_m_n.GetLength(I1) &&
+                K0 == b_grid_desc_b_k0_n_k1.GetLength(I1) &&
+                K1 == a_grid_desc_b_k0_m_k1.GetLength(I3) &&
+                K1 == b_grid_desc_b_k0_n_k1.GetLength(I3)) &&
+               KBatch == b_grid_desc_b_k0_n_k1.GetLength(I0) &&
+               (M % MPerBlock == 0 && N % NPerBlock == 0 && K0 % K0PerBlock == 0);
+    }
+
+    __host__ __device__ static constexpr index_t CalculateGridSize(index_t M, index_t N)
+    {
+        const index_t grid_size = (M / MPerBlock) * (N / NPerBlock);
+
+        return grid_size;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K0)
+    {
+        const bool has_main_k_block_loop = (K0 + K0PerBlock) / (2 * K0PerBlock) > 1;
+
+        return has_main_k_block_loop;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasDoubleTailKBlockLoop(index_t K0)
+    {
+        const bool has_double_tail_k_block_loop = (K0 / K0PerBlock) % 2 == 0;
+
+        return has_double_tail_k_block_loop;
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeAGridDescriptor_B_K0_M0_M1_K1(const AGridDesc_B_K0_M_K1& a_grid_desc_b_k0_m_k1)
+    {
+        const auto KBatch = a_grid_desc_b_k0_m_k1.GetLength(I0);
+        const auto K0     = a_grid_desc_b_k0_m_k1.GetLength(I1);
+        const auto M      = a_grid_desc_b_k0_m_k1.GetLength(I2);
+
+        const auto M1 = Number<MPerBlock>{};
+        const auto M0 = M / M1;
+
+        const auto a_grid_desc_b_k0_m0_m1_k1 = transform_tensor_descriptor(
+            a_grid_desc_b_k0_m_k1,
+            make_tuple(make_pass_through_transform(KBatch),
+                       make_pass_through_transform(K0),
+                       make_unmerge_transform(make_tuple(M0, M1)),
+                       make_pass_through_transform(K1)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
+
+        return a_grid_desc_b_k0_m0_m1_k1;
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeBGridDescriptor_B_K0_N0_N1_K1(const BGridDesc_B_K0_N_K1& b_grid_desc_b_k0_n_k1)
+    {
+        const auto KBatch = b_grid_desc_b_k0_n_k1.GetLength(I0);
+        const auto K0     = b_grid_desc_b_k0_n_k1.GetLength(I1);
+        const auto N      = b_grid_desc_b_k0_n_k1.GetLength(I2);
+
+        const auto N1 = Number<NPerBlock>{};
+        const auto N0 = N / N1;
+
+        const auto b_grid_desc_b_k0_n0_n1_k1 = transform_tensor_descriptor(
+            b_grid_desc_b_k0_n_k1,
+            make_tuple(make_pass_through_transform(KBatch),
+                       make_pass_through_transform(K0),
+                       make_unmerge_transform(make_tuple(N0, N1)),
+                       make_pass_through_transform(K1)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
+
+        return b_grid_desc_b_k0_n0_n1_k1;
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        const auto M = c_grid_desc_m_n.GetLength(I0);
+        const auto N = c_grid_desc_m_n.GetLength(I1);
+
+        constexpr auto M1 = Number<MPerBlock>{};
+        constexpr auto N1 = Number<NPerBlock>{};
+
+        const auto M0 = M / M1;
+        const auto N0 = N / N1;
+
+        constexpr auto M11 =
+            Number<container_reduce(M11N11ThreadClusterM110Xs{}, math::multiplies{}, I1) *
+                   M1PerThreadM111>{};
+        constexpr auto N11 =
+            Number<container_reduce(M11N11ThreadClusterN110Xs{}, math::multiplies{}, I1) *
+                   N1PerThreadN111>{};
+
+        constexpr auto M10 = M1 / M11;
+        constexpr auto N10 = N1 / N11;
+
+        const auto c_grid_desc_m0_m10_m11_n0_n10_n11 = transform_tensor_descriptor(
+            c_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(M0, M10, M11)),
+                       make_unmerge_transform(make_tuple(N0, N10, N11))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1, 2>{}, Sequence<3, 4, 5>{}));
+
+        return c_grid_desc_m0_m10_m11_n0_n10_n11;
+    }
+
+    // return block_id to C matrix tile idx (m0, n0) mapping
+    __host__ __device__ static constexpr auto MakeCBlockClusterAdaptor(
+        const CGridDesc_M_N& c_m_n_grid_desc, index_t M01, index_t N01, index_t KBatch)
+    {
+        return BlockToCTileMap_KSplit_M00_N00_M01_N01<MPerBlock, NPerBlock, CGridDesc_M_N>(
+            c_m_n_grid_desc, M01, N01, KBatch);
+    }
+
+    using AGridDesc_B_K0_M0_M1_K1 =
+        decltype(MakeAGridDescriptor_B_K0_M0_M1_K1(AGridDesc_B_K0_M_K1{}));
+    using BGridDesc_B_K0_N0_N1_K1 =
+        decltype(MakeBGridDescriptor_B_K0_N0_N1_K1(BGridDesc_B_K0_N_K1{}));
+    using CGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(CGridDesc_M_N{}));
+    using CBlockClusterAdaptor = decltype(MakeCBlockClusterAdaptor(CGridDesc_M_N{}, 1, 1, 1));
+
+    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
+    __device__ static void
+    Run(const FloatAB* __restrict__ p_a_grid,
+        const FloatAB* __restrict__ p_b_grid,
+        FloatC* __restrict__ p_c_grid,
+        FloatAB* __restrict__ p_shared_block,
+        const AGridDesc_B_K0_M0_M1_K1& a_grid_desc_b_k0_m0_m1_k1,
+        const BGridDesc_B_K0_N0_N1_K1& b_grid_desc_b_k0_n0_n1_k1,
+        const CGridDesc_M0_M10_M11_N0_N10_N11& c_grid_desc_m0_m10_m11_n0_n10_n11,
+        const CBlockClusterAdaptor& c_block_cluster_adaptor,
+        integral_constant<bool, HasMainKBlockLoop>,
+        integral_constant<bool, HasDoubleTailKBlockLoop>)
+    {
+        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_b_k0_m0_m1_k1.GetElementSpaceSize());
+        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_b_k0_n0_n1_k1.GetElementSpaceSize());
+        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_c_grid, c_grid_desc_m0_m10_m11_n0_n10_n11.GetElementSpaceSize());
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            c_block_cluster_adaptor.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        const index_t k_batch_id = block_work_idx[I0];
+
+        if(!c_block_cluster_adaptor.ValidCTileIndex(
+               make_tuple(block_work_idx[I1], block_work_idx[I2]),
+               make_tuple(c_grid_desc_m0_m10_m11_n0_n10_n11.GetLength(I0),
+                          c_grid_desc_m0_m10_m11_n0_n10_n11.GetLength(I3))))
+        {
+            return;
+        }
+
+        // HACK: this force m/n_block_data_idx_on_grid into SGPR
+        const index_t m_block_data_idx_on_grid = __builtin_amdgcn_readfirstlane(block_work_idx[I1]);
+
+        const index_t n_block_data_idx_on_grid = __builtin_amdgcn_readfirstlane(block_work_idx[I2]);
+
+        // TODO: change this. I think it needs multi-dimensional alignment
+        constexpr auto max_lds_align = K1;
+
+        // TODO: check alignment
+        // A matrix in LDS memory, dst of blockwise copy
+        //   be careful of LDS alignment
+        constexpr auto a_block_desc_b_k0_m0_m1_k1 = make_naive_tensor_descriptor_aligned(
+            make_tuple(I1, Number<K0PerBlock>{}, I1, Number<MPerBlock>{}, K1), max_lds_align);
+
+        // TODO: check alignment
+        // B matrix in LDS memory, dst of blockwise copy
+        //   be careful of LDS alignment
+        constexpr auto b_block_desc_b_k0_n0_n1_k1 = make_naive_tensor_descriptor_aligned(
+            make_tuple(I1, Number<K0PerBlock>{}, I1, Number<NPerBlock>{}, K1), max_lds_align);
+
+        // TODO: check alignment
+        // A matrix in LDS memory, dst of blockwise copy
+        //   be careful of LDS alignment
+        constexpr auto a_block_desc_k0_m0_m1_k1 = make_naive_tensor_descriptor_aligned(
+            make_tuple(Number<K0PerBlock>{}, I1, Number<MPerBlock>{}, K1), max_lds_align);
+
+        // TODO: check alignment
+        // B matrix in LDS memory, dst of blockwise copy
+        //   be careful of LDS alignment
+        constexpr auto b_block_desc_k0_n0_n1_k1 = make_naive_tensor_descriptor_aligned(
+            make_tuple(Number<K0PerBlock>{}, I1, Number<NPerBlock>{}, K1), max_lds_align);
+
+        // TODO: check alignment
+        // A matrix in LDS memory, for blockwise GEMM
+        constexpr auto a_k0_m_k1_block_desc = make_naive_tensor_descriptor_aligned(
+            make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1), max_lds_align);
+
+        // TODO: check alignment
+        // B matrix in LDS memory, for blockwise GEMM
+        constexpr auto b_k0_n_k1_block_desc = make_naive_tensor_descriptor_aligned(
+            make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1), max_lds_align);
+
+        static_assert(a_block_desc_k0_m0_m1_k1.GetElementSpaceSize() ==
+                          a_k0_m_k1_block_desc.GetElementSpaceSize() &&
+                      b_block_desc_k0_n0_n1_k1.GetElementSpaceSize() ==
+                          b_k0_n_k1_block_desc.GetElementSpaceSize() &&
+                      "wrong!");
+
+        // A matrix blockwise copy
+        auto a_blockwise_copy = BlockwiseTensorSliceTransfer_v5r1<
+            BlockSize,
+            InMemoryDataOperationEnum::Set,
+            Sequence<1, K0PerBlock, 1, MPerBlock, K1.value>,
+            ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+            ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+            ABlockTransferThreadClusterArrangeOrder,
+            FloatAB,
+            FloatAB,
+            remove_reference_t<decltype(a_grid_desc_b_k0_m0_m1_k1)>,
+            decltype(a_block_desc_b_k0_m0_m1_k1),
+            ABlockTransferSrcAccessOrder,
+            Sequence<0, 1, 2, 3, 4>,
+            ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1, // SrcVectorTensorLengths
+            ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1, // DstVectorTensorLengths
+            ABlockTransferSrcVectorTensorContiguousDimOrder,  // SrcVectorTensorContiguousDimOrder
+            Sequence<0, 1, 2, 3, 4>,                          // DstVectorTensorContiguousDimOrder
+            false,
+            true>(a_grid_desc_b_k0_m0_m1_k1,
+                  make_multi_index(k_batch_id, 0, m_block_data_idx_on_grid, 0, 0),
+                  a_block_desc_b_k0_m0_m1_k1,
+                  make_multi_index(0, 0, 0, 0, 0));
+
+        // B matrix blockwise copy
+        auto b_blockwise_copy = BlockwiseTensorSliceTransfer_v5r1<
+            BlockSize,
+            InMemoryDataOperationEnum::Set,
+            Sequence<1, K0PerBlock, 1, NPerBlock, K1.value>,
+            BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+            BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+            BBlockTransferThreadClusterArrangeOrder,
+            FloatAB,
+            FloatAB,
+            remove_reference_t<decltype(b_grid_desc_b_k0_n0_n1_k1)>,
+            decltype(b_block_desc_b_k0_n0_n1_k1),
+            BBlockTransferSrcAccessOrder,
+            Sequence<0, 1, 2, 3, 4>,
+            BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1, // SrcVectorTensorLengths
+            BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1, // DstVectorTensorLengths
+            BBlockTransferSrcVectorTensorContiguousDimOrder,  // SrcVectorTensorContiguousDimOrder
+            Sequence<0, 1, 2, 3, 4>,                          // DstVectorTensorContiguousDimOrder
+            false,
+            true>(b_grid_desc_b_k0_n0_n1_k1,
+                  make_multi_index(k_batch_id, 0, n_block_data_idx_on_grid, 0, 0),
+                  b_block_desc_b_k0_n0_n1_k1,
+                  make_multi_index(0, 0, 0, 0, 0));
+
+        // GEMM definition
+        //   c_mtx += transpose(a_mtx) * b_mtx
+        //     a_mtx[K0PerBlock, MPerBlock] is in LDS
+        //     b_mtx[KPerBlocl, NPerBlock] is in LDS
+        //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
+        //       register
+        const auto blockwise_gemm =
+            BlockwiseGemmDl_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_pipeline_BM0_2_BN0_2<
+                BlockSize,
+                FloatAB,
+                FloatAB,
+                FloatAcc,
+                decltype(a_k0_m_k1_block_desc),
+                decltype(b_k0_n_k1_block_desc),
+                M1PerThreadM111,
+                N1PerThreadN111,
+                KPerThread,
+                M11N11ThreadClusterM110Xs,
+                M11N11ThreadClusterN110Xs,
+                M1PerThreadM111,
+                N1PerThreadN111>{};
+
+        constexpr auto c_m10_m11_n10_n11_thread_tensor_lengths =
+            decltype(blockwise_gemm)::GetCThreadTensorLengths_BM0_BM1_BN0_BN1();
+
+        constexpr auto c_thread_desc_m10_m11_n10_n11 = make_naive_tensor_descriptor_packed(
+            sequence_to_tuple_of_number(c_m10_m11_n10_n11_thread_tensor_lengths));
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_aligned_space_size = math::integer_least_multiple(
+            a_block_desc_k0_m0_m1_k1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_aligned_space_size = math::integer_least_multiple(
+            b_block_desc_k0_n0_n1_k1.GetElementSpaceSize(), max_lds_align);
+
+        FloatAB* p_a_block_double = p_shared_block;
+        FloatAB* p_b_block_double = p_shared_block + 2 * a_block_aligned_space_size;
+
+        // register allocation for output
+        auto c_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAcc>(
+            c_thread_desc_m10_m11_n10_n11.GetElementSpaceSize());
+
+        // Initialize C
+        c_thread_buf.Clear();
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0, 0);
+
+        auto a_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            p_a_block_double, a_block_desc_k0_m0_m1_k1.GetElementSpaceSize());
+        auto b_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            p_b_block_double, b_block_desc_k0_n0_n1_k1.GetElementSpaceSize());
+
+        auto a_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            p_a_block_double + a_block_aligned_space_size,
+            a_block_desc_k0_m0_m1_k1.GetElementSpaceSize());
+        auto b_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            p_b_block_double + b_block_aligned_space_size,
+            b_block_desc_k0_n0_n1_k1.GetElementSpaceSize());
+
+        // LDS double buffer: preload data into LDS
+        {
+            a_blockwise_copy.RunRead(a_grid_desc_b_k0_m0_m1_k1, a_global_buf);
+            b_blockwise_copy.RunRead(b_grid_desc_b_k0_n0_n1_k1, b_global_buf);
+
+            a_blockwise_copy.RunWrite(a_block_desc_b_k0_m0_m1_k1, a_block_even_buf);
+            b_blockwise_copy.RunWrite(b_block_desc_b_k0_n0_n1_k1, b_block_even_buf);
+        }
+
+        if constexpr(HasMainKBlockLoop)
+        {
+            const auto K0 = a_grid_desc_b_k0_m0_m1_k1.GetLength(I1);
+
+            index_t k_block_data_begin = 0;
+
+            // LDS double buffer: main body
+            // use Do-While loop instead of For loop to simplify control flow
+            do
+            {
+                // even iteration
+                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_b_k0_m0_m1_k1,
+                                                    a_block_slice_copy_step);
+                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_b_k0_n0_n1_k1,
+                                                    b_block_slice_copy_step);
+
+                // LDS doubel buffer: load next data from device mem
+                a_blockwise_copy.RunRead(a_grid_desc_b_k0_m0_m1_k1, a_global_buf);
+                b_blockwise_copy.RunRead(b_grid_desc_b_k0_n0_n1_k1, b_global_buf);
+
+                block_sync_lds();
+
+                // LDS double buffer: GEMM on current data
+                blockwise_gemm.Run(c_thread_desc_m10_m11_n10_n11,
+                                   a_block_even_buf,
+                                   b_block_even_buf,
+                                   c_thread_buf);
+
+                // LDS double buffer: store next data to LDS
+                a_blockwise_copy.RunWrite(a_block_desc_b_k0_m0_m1_k1, a_block_odd_buf);
+                b_blockwise_copy.RunWrite(b_block_desc_b_k0_n0_n1_k1, b_block_odd_buf);
+
+                // odd iteration
+                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_b_k0_m0_m1_k1,
+                                                    a_block_slice_copy_step);
+                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_b_k0_n0_n1_k1,
+                                                    b_block_slice_copy_step);
+
+                // LDS doubel buffer: load next data from device mem
+                a_blockwise_copy.RunRead(a_grid_desc_b_k0_m0_m1_k1, a_global_buf);
+                b_blockwise_copy.RunRead(b_grid_desc_b_k0_n0_n1_k1, b_global_buf);
+
+                block_sync_lds();
+
+                // LDS double buffer: GEMM on current data
+                blockwise_gemm.Run(
+                    c_thread_desc_m10_m11_n10_n11, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
+
+                // LDS double buffer: store next data to LDS
+                a_blockwise_copy.RunWrite(a_block_desc_b_k0_m0_m1_k1, a_block_even_buf);
+                b_blockwise_copy.RunWrite(b_block_desc_b_k0_n0_n1_k1, b_block_even_buf);
+
+                k_block_data_begin += 2 * K0PerBlock;
+            } while(k_block_data_begin < K0 - 2 * K0PerBlock);
+        }
+
+        // LDS double buffer: tail
+        if constexpr(HasDoubleTailKBlockLoop) // if has 2 iteration left
+        {
+            a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_b_k0_m0_m1_k1, a_block_slice_copy_step);
+            b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_b_k0_n0_n1_k1, b_block_slice_copy_step);
+
+            block_sync_lds();
+
+            // LDS double buffer: load last data from device mem
+            a_blockwise_copy.RunRead(a_grid_desc_b_k0_m0_m1_k1, a_global_buf);
+            b_blockwise_copy.RunRead(b_grid_desc_b_k0_n0_n1_k1, b_global_buf);
+
+            // LDS double buffer: GEMM on 2nd-last data
+            blockwise_gemm.Run(
+                c_thread_desc_m10_m11_n10_n11, a_block_even_buf, b_block_even_buf, c_thread_buf);
+
+            // LDS double buffer: store last data to LDS
+            a_blockwise_copy.RunWrite(a_block_desc_b_k0_m0_m1_k1, a_block_odd_buf);
+            b_blockwise_copy.RunWrite(b_block_desc_b_k0_n0_n1_k1, b_block_odd_buf);
+
+            block_sync_lds();
+
+            // LDS double buffer: GEMM on last data
+            blockwise_gemm.Run(
+                c_thread_desc_m10_m11_n10_n11, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
+        }
+        else // if has 1 iteration left
+        {
+            __syncthreads();
+
+            // LDS double buffer: GEMM on last data
+            blockwise_gemm.Run(
+                c_thread_desc_m10_m11_n10_n11, a_block_even_buf, b_block_even_buf, c_thread_buf);
+        }
+
+        // output: register to global memory
+        {
+            constexpr auto c_thread_desc_m0_m10_m11_n0_n10_n11 =
+                make_naive_tensor_descriptor_packed(
+                    make_tuple(I1,
+                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I0]>{},
+                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I1]>{},
+                               I1,
+                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I2]>{},
+                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I3]>{}));
+
+            const auto c_m10_m11_n10_n11_thread_origin_idx_on_block =
+                blockwise_gemm.CalculateCThreadOriginOnBlock_BM0_BM1_BN0_BN1(
+                    get_thread_local_1d_id());
+
+            ThreadwiseTensorSliceTransfer_v1r3<
+                FloatAcc,
+                FloatC,
+                decltype(c_thread_desc_m0_m10_m11_n0_n10_n11),
+                decltype(c_grid_desc_m0_m10_m11_n0_n10_n11),
+                ck::tensor_operation::element_wise::PassThrough,
+                Sequence<1,
+                         c_m10_m11_n10_n11_thread_tensor_lengths[I0],
+                         c_m10_m11_n10_n11_thread_tensor_lengths[I1],
+                         1,
+                         c_m10_m11_n10_n11_thread_tensor_lengths[I2],
+                         c_m10_m11_n10_n11_thread_tensor_lengths[I3]>,
+                CThreadTransferSrcDstAccessOrder,
+                CThreadTransferSrcDstVectorDim,
+                CThreadTransferDstScalarPerVector,
+                CGlobalMemoryDataOperation,
+                1,
+                true>{c_grid_desc_m0_m10_m11_n0_n10_n11,
+                      make_multi_index(m_block_data_idx_on_grid,
+                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I0],
+                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I1],
+                                       n_block_data_idx_on_grid,
+                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I2],
+                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I3]),
+                      ck::tensor_operation::element_wise::PassThrough{}}
+                .Run(c_thread_desc_m0_m10_m11_n0_n10_n11,
+                     make_tuple(I0, I0, I0, I0, I0, I0),
+                     c_thread_buf,
+                     c_grid_desc_m0_m10_m11_n0_n10_n11,
+                     c_grid_buf);
+        }
+    }
+};
+
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v1r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v1r2.hpp
deleted file mode 100644
index 84e033e1e91bb9873ebed66906c6da5a287bf1bd..0000000000000000000000000000000000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v1r2.hpp
+++ /dev/null
@@ -1,608 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_GEMM_DLOPS_V1R2_HPP
-#define CK_GRIDWISE_GEMM_DLOPS_V1R2_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_gemm_dlops_v2r2.hpp"
-#include "blockwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_set.hpp"
-
-namespace ck {
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AKM0M1GridDesc,
-          typename BKN0N1GridDesc,
-          typename CM0M10M11N0N10N11GridDesc,
-          typename CBlockIdToM0N0BlockClusterAdaptor,
-          bool HasMainKBlockLoop,
-          bool HasDoubleTailKBlockLoop>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v1r2(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AKM0M1GridDesc a_k_m0_m1_grid_desc,
-            const BKN0N1GridDesc b_k_n0_n1_grid_desc,
-            const CM0M10M11N0N10N11GridDesc c_m0_m10_m11_n0_n10_n11_grid_desc,
-            const CBlockIdToM0N0BlockClusterAdaptor cblockid_to_m0_n0_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::Run(p_a_grid,
-                      p_b_grid,
-                      p_c_grid,
-                      p_shared_block,
-                      a_k_m0_m1_grid_desc,
-                      b_k_n0_n1_grid_desc,
-                      c_m0_m10_m11_n0_n10_n11_grid_desc,
-                      cblockid_to_m0_n0_block_cluster_adaptor,
-                      integral_constant<bool, HasMainKBlockLoop>{},
-                      integral_constant<bool, HasDoubleTailKBlockLoop>{});
-}
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AKMGridDesc,
-          typename BKNGridDesc,
-          typename CMNGridDesc,
-          index_t MPerBlockM1,
-          index_t NPerBlockN1,
-          index_t KPerBlock,
-          index_t M1PerThreadM111,
-          index_t N1PerThreadN111,
-          index_t KPerThread,
-          index_t M11N11ThreadClusterM1100,
-          index_t M11N11ThreadClusterN1100,
-          index_t M11N11ThreadClusterM1101,
-          index_t M11N11ThreadClusterN1101,
-          typename ABlockTransferThreadSliceLengths_K_M0_M1,
-          typename ABlockTransferThreadClusterLengths_K_M0_M1,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_M1,
-          bool AThreadTransferSrcResetCoordinateAfterRun,
-          typename BBlockTransferThreadSliceLengths_K_N0_N1,
-          typename BBlockTransferThreadClusterLengths_K_N0_N1,
-          typename BBlockTransferThreadClusterArrangeOrder,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          index_t BBlockTransferDstScalarPerVector_N1,
-          bool BThreadTransferSrcResetCoordinateAfterRun,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGridStepHacks,
-          typename BGridStepHacks,
-          typename CGridStepHacks,
-          typename AGridMoveSliceWindowStepHacks,
-          typename BGridMoveSliceWindowStepHacks>
-struct GridwiseGemmDlops_km_kn_mn_v1r2
-{
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        constexpr auto max_lds_align = math::lcm(Number<ABlockTransferDstScalarPerVector_M1>{},
-                                                 Number<BBlockTransferDstScalarPerVector_N1>{},
-                                                 Number<M1PerThreadM111>{},
-                                                 Number<N1PerThreadN111>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_k_m_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<MPerBlockM1>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_k_n_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<NPerBlockN1>{}), max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size =
-            math::integer_least_multiple(a_k_m_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size =
-            math::integer_least_multiple(b_k_n_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        return 2 * (a_block_aligned_space_size + b_block_aligned_space_size) * sizeof(FloatAB);
-    }
-
-    __host__ __device__ static constexpr bool CheckValidity(const AKMGridDesc& a_k_m_grid_desc,
-                                                            const BKNGridDesc& b_k_n_grid_desc,
-                                                            const CMNGridDesc& c_m_n_grid_desc)
-    {
-        const auto M = a_k_m_grid_desc.GetLength(I1);
-        const auto N = b_k_n_grid_desc.GetLength(I1);
-        const auto K = a_k_m_grid_desc.GetLength(I0);
-
-        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
-
-        return (M == c_m_n_grid_desc.GetLength(I0) && N == c_m_n_grid_desc.GetLength(I1) &&
-                K == b_k_n_grid_desc.GetLength(I0)) &&
-               (M % MPerBlockM1 == 0 && N % NPerBlockN1 == 0 && K % KPerBlock == 0);
-    }
-
-    __host__ __device__ static constexpr index_t CalculateGridSize(index_t M, index_t N)
-    {
-        const index_t grid_size = (M / MPerBlockM1) * (N / NPerBlockN1);
-
-        return grid_size;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
-    {
-        const bool has_main_k_block_loop = (K + KPerBlock) / (2 * KPerBlock) > 1;
-
-        return has_main_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasDoubleTailKBlockLoop(index_t K)
-    {
-        const bool has_double_tail_k_block_loop = (K / KPerBlock) % 2 == 0;
-
-        return has_double_tail_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeAKM0M1GridDescriptor(const AKMGridDesc& a_k_m_grid_desc)
-    {
-        const auto K = a_k_m_grid_desc.GetLength(I0);
-        const auto M = a_k_m_grid_desc.GetLength(I1);
-
-        const auto M1 = Number<MPerBlockM1>{};
-        const auto M0 = M / M1;
-
-        const auto a_k_m0_m1_grid_desc = transform_tensor_descriptor(
-            a_k_m_grid_desc,
-            make_tuple(make_pass_through_transform(K), make_unmerge_transform(make_tuple(M0, M1))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}));
-
-        return a_k_m0_m1_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeBKN0N1GridDescriptor(const BKNGridDesc& b_k_n_grid_desc)
-    {
-        const auto K = b_k_n_grid_desc.GetLength(I0);
-        const auto N = b_k_n_grid_desc.GetLength(I1);
-
-        const auto N1 = Number<NPerBlockN1>{};
-        const auto N0 = N / N1;
-
-        const auto b_k_n0_n1_grid_desc = transform_tensor_descriptor(
-            b_k_n_grid_desc,
-            make_tuple(make_pass_through_transform(K), make_unmerge_transform(make_tuple(N0, N1))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}));
-
-        return b_k_n0_n1_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCM0M10M11N0N10N11GridDescriptor(const CMNGridDesc& c_m_n_grid_desc)
-    {
-        const auto M = c_m_n_grid_desc.GetLength(I0);
-        const auto N = c_m_n_grid_desc.GetLength(I1);
-
-        constexpr auto M1 = Number<MPerBlockM1>{};
-        constexpr auto N1 = Number<NPerBlockN1>{};
-
-        const auto M0 = M / M1;
-        const auto N0 = N / N1;
-
-        constexpr auto M11 =
-            Number<M11N11ThreadClusterM1100 * M11N11ThreadClusterM1101 * M1PerThreadM111>{};
-        constexpr auto N11 =
-            Number<M11N11ThreadClusterN1100 * M11N11ThreadClusterN1101 * N1PerThreadN111>{};
-
-        constexpr auto M10 = M1 / M11;
-        constexpr auto N10 = N1 / N11;
-
-        const auto c_m0_m10_m11_n0_n10_n11_grid_desc = transform_tensor_descriptor(
-            c_m_n_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(M0, M10, M11)),
-                       make_unmerge_transform(make_tuple(N0, N10, N11))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0, 1, 2>{}, Sequence<3, 4, 5>{}));
-
-        return c_m0_m10_m11_n0_n10_n11_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCBlockIdToM0N0BlockClusterAdaptor(const CMNGridDesc& c_m_n_grid_desc)
-    {
-        const auto M = c_m_n_grid_desc.GetLength(I0);
-        const auto N = c_m_n_grid_desc.GetLength(I1);
-
-        constexpr auto M1 = Number<MPerBlockM1>{};
-        constexpr auto N1 = Number<NPerBlockN1>{};
-
-        const auto M0 = M / M1;
-        const auto N0 = N / N1;
-
-        const auto cblockid_to_m0_n0_block_cluster_adaptor =
-            make_single_stage_tensor_adaptor(make_tuple(make_merge_transform(make_tuple(M0, N0))),
-                                             make_tuple(Sequence<0, 1>{}),
-                                             make_tuple(Sequence<0>{}));
-
-        return cblockid_to_m0_n0_block_cluster_adaptor;
-    }
-
-    using AKM0M1GridDesc            = decltype(MakeAKM0M1GridDescriptor(AKMGridDesc{}));
-    using BKN0N1GridDesc            = decltype(MakeBKN0N1GridDescriptor(BKNGridDesc{}));
-    using CM0M10M11N0N10N11GridDesc = decltype(MakeCM0M10M11N0N10N11GridDescriptor(CMNGridDesc{}));
-    using CBlockIdToM0N0BlockClusterAdaptor =
-        decltype(MakeCBlockIdToM0N0BlockClusterAdaptor(CMNGridDesc{}));
-
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ static void
-    Run(const FloatAB* __restrict__ p_a_grid,
-        const FloatAB* __restrict__ p_b_grid,
-        FloatC* __restrict__ p_c_grid,
-        FloatAB* __restrict__ p_shared_block,
-        const AKM0M1GridDesc& a_k_m0_m1_grid_desc,
-        const BKN0N1GridDesc& b_k_n0_n1_grid_desc,
-        const CM0M10M11N0N10N11GridDesc& c_m0_m10_m11_n0_n10_n11_grid_desc,
-        const CBlockIdToM0N0BlockClusterAdaptor& cblockid_to_m0_n0_block_cluster_adaptor,
-        integral_constant<bool, HasMainKBlockLoop>,
-        integral_constant<bool, HasDoubleTailKBlockLoop>)
-    {
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_grid, a_k_m0_m1_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_grid, b_k_n0_n1_grid_desc.GetElementSpaceSize());
-        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_grid, c_m0_m10_m11_n0_n10_n11_grid_desc.GetElementSpaceSize());
-
-        const auto K = a_k_m0_m1_grid_desc.GetLength(I0);
-
-        // divide block work by [M, N]
-        const auto c_m0_n0_block_cluster_idx =
-            cblockid_to_m0_n0_block_cluster_adaptor.CalculateBottomIndex(
-                make_multi_index(get_block_1d_id()));
-
-        // HACK: this force index data into SGPR
-        const index_t im0 = __builtin_amdgcn_readfirstlane(c_m0_n0_block_cluster_idx[I0]);
-        const index_t in0 = __builtin_amdgcn_readfirstlane(c_m0_n0_block_cluster_idx[I1]);
-
-        // lds max alignment
-        constexpr auto max_lds_align = math::lcm(Number<ABlockTransferDstScalarPerVector_M1>{},
-                                                 Number<BBlockTransferDstScalarPerVector_N1>{},
-                                                 Number<M1PerThreadM111>{},
-                                                 Number<N1PerThreadN111>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_k_m_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<MPerBlockM1>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_k_n_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<NPerBlockN1>{}), max_lds_align);
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_k_m0_m1_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, I1, Number<MPerBlockM1>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_k_n0_n1_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, I1, Number<NPerBlockN1>{}), max_lds_align);
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<KPerBlock, 1, MPerBlockM1>,
-                                            ABlockTransferThreadSliceLengths_K_M0_M1,
-                                            ABlockTransferThreadClusterLengths_K_M0_M1,
-                                            ABlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(a_k_m0_m1_grid_desc),
-                                            decltype(a_k_m0_m1_block_desc),
-                                            ABlockTransferSrcAccessOrder,
-                                            Sequence<0, 1, 2>,
-                                            ABlockTransferSrcVectorDim,
-                                            2,
-                                            ABlockTransferSrcScalarPerVector,
-                                            ABlockTransferDstScalarPerVector_M1,
-                                            1,
-                                            1,
-                                            AThreadTransferSrcResetCoordinateAfterRun,
-                                            true>(a_k_m0_m1_grid_desc,
-                                                  make_multi_index(0, im0, 0),
-                                                  a_k_m0_m1_block_desc,
-                                                  make_multi_index(0, 0, 0));
-
-        // B matrix blockwise copy
-        auto b_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<KPerBlock, 1, NPerBlockN1>,
-                                            BBlockTransferThreadSliceLengths_K_N0_N1,
-                                            BBlockTransferThreadClusterLengths_K_N0_N1,
-                                            BBlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(b_k_n0_n1_grid_desc),
-                                            decltype(b_k_n0_n1_block_desc),
-                                            BBlockTransferSrcAccessOrder,
-                                            Sequence<0, 1, 2>,
-                                            BBlockTransferSrcVectorDim,
-                                            2,
-                                            BBlockTransferSrcScalarPerVector,
-                                            BBlockTransferDstScalarPerVector_N1,
-                                            1,
-                                            1,
-                                            BThreadTransferSrcResetCoordinateAfterRun,
-                                            true>(b_k_n0_n1_grid_desc,
-                                                  make_multi_index(0, in0, 0),
-                                                  b_k_n0_n1_block_desc,
-                                                  make_multi_index(0, 0, 0));
-
-        // GEMM definition
-        //   c_mtx += transpose(a_mtx) * b_mtx
-        //     a_mtx[KPerBlock, MPerBlockM1] is in LDS
-        //     b_mtx[KPerBlocl, NPerBlockN1] is in LDS
-        //     c_mtx[MPerBlockM1, NPerBlockN1] is distributed among threads, and saved in
-        //       register
-        const auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v2r2_pipeline_2x2<BlockSize,
-                                                                FloatAB,
-                                                                FloatAB,
-                                                                FloatAcc,
-                                                                decltype(a_k_m_block_desc),
-                                                                decltype(b_k_n_block_desc),
-                                                                M1PerThreadM111,
-                                                                N1PerThreadN111,
-                                                                KPerThread,
-                                                                M11N11ThreadClusterM1100,
-                                                                M11N11ThreadClusterN1100,
-                                                                M11N11ThreadClusterM1101,
-                                                                M11N11ThreadClusterN1101,
-                                                                M1PerThreadM111,
-                                                                N1PerThreadN111>{};
-        constexpr auto c_m10_m11_n10_n11_thread_tensor_lengths =
-            decltype(blockwise_gemm)::GetCM0M1N0N1ThreadTensorLengths();
-
-        constexpr auto c_m10_m11_n10_n11_thread_desc = make_naive_tensor_descriptor_packed(
-            sequence_to_tuple_of_number(c_m10_m11_n10_n11_thread_tensor_lengths));
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size =
-            math::integer_least_multiple(a_k_m0_m1_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size =
-            math::integer_least_multiple(b_k_n0_n1_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        FloatAB* p_a_block_double = p_shared_block;
-        FloatAB* p_b_block_double = p_shared_block + 2 * a_block_aligned_space_size;
-
-        // register allocation for output
-        auto c_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAcc>(
-            c_m10_m11_n10_n11_thread_desc.GetElementSpaceSize());
-
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_m10_m11_n10_n11_thread_desc),
-                                    decltype(c_m10_m11_n10_n11_thread_tensor_lengths)>{}
-            .Run(c_m10_m11_n10_n11_thread_desc,
-                 make_tuple(I0, I0, I0, I0),
-                 c_thread_buf,
-                 FloatAcc{0});
-
-        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock, 0, 0);
-        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock, 0, 0);
-
-        // hack to control index calculation when iterating over A and B matrix for threadwise copy
-        constexpr auto a_k_m0_m1_global_step_hacks = AGridStepHacks{};
-        constexpr auto b_k_n0_n1_global_step_hacks = BGridStepHacks{};
-
-        // hack to control index calculation when move slice window for A and B matrix for
-        // threadwise copy
-        constexpr auto a_k_m0_m1_global_move_slice_window_step_hack =
-            AGridMoveSliceWindowStepHacks{};
-        constexpr auto b_k_n0_n1_global_move_slice_window_step_hack =
-            BGridMoveSliceWindowStepHacks{};
-
-        auto a_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double, a_k_m0_m1_block_desc.GetElementSpaceSize());
-        auto b_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double, b_k_n0_n1_block_desc.GetElementSpaceSize());
-
-        auto a_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double + a_block_aligned_space_size,
-            a_k_m0_m1_block_desc.GetElementSpaceSize());
-        auto b_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double + b_block_aligned_space_size,
-            b_k_n0_n1_block_desc.GetElementSpaceSize());
-
-        // LDS double buffer: preload data into LDS
-        {
-            a_blockwise_copy.RunRead(
-                a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-            b_blockwise_copy.RunRead(
-                b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-            a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_even_buf);
-            b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_even_buf);
-        }
-
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t k_block_data_begin = 0;
-
-            // LDS double buffer: main body
-            // use Do-While loop instead of For loop to simplify control flow
-            do
-            {
-                // even iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_k_m0_m1_grid_desc,
-                                                    a_block_slice_copy_step,
-                                                    a_k_m0_m1_global_move_slice_window_step_hack);
-                b_blockwise_copy.MoveSrcSliceWindow(b_k_n0_n1_grid_desc,
-                                                    b_block_slice_copy_step,
-                                                    b_k_n0_n1_global_move_slice_window_step_hack);
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-                b_blockwise_copy.RunRead(
-                    b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(c_m10_m11_n10_n11_thread_desc,
-                                   a_block_even_buf,
-                                   b_block_even_buf,
-                                   c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_odd_buf);
-                b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_odd_buf);
-
-                // odd iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_k_m0_m1_grid_desc,
-                                                    a_block_slice_copy_step,
-                                                    a_k_m0_m1_global_move_slice_window_step_hack);
-                b_blockwise_copy.MoveSrcSliceWindow(b_k_n0_n1_grid_desc,
-                                                    b_block_slice_copy_step,
-                                                    b_k_n0_n1_global_move_slice_window_step_hack);
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-                b_blockwise_copy.RunRead(
-                    b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(
-                    c_m10_m11_n10_n11_thread_desc, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_even_buf);
-                b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_even_buf);
-
-                k_block_data_begin += 2 * KPerBlock;
-            } while(k_block_data_begin < K - 2 * KPerBlock);
-        }
-
-        // LDS double buffer: tail
-        if constexpr(HasDoubleTailKBlockLoop) // if has 2 iteration left
-        {
-            a_blockwise_copy.MoveSrcSliceWindow(a_k_m0_m1_grid_desc,
-                                                a_block_slice_copy_step,
-                                                a_k_m0_m1_global_move_slice_window_step_hack);
-            b_blockwise_copy.MoveSrcSliceWindow(b_k_n0_n1_grid_desc,
-                                                b_block_slice_copy_step,
-                                                b_k_n0_n1_global_move_slice_window_step_hack);
-
-            __syncthreads();
-
-            // LDS double buffer: load last data from device mem
-            a_blockwise_copy.RunRead(
-                a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-            b_blockwise_copy.RunRead(
-                b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-            // LDS double buffer: GEMM on 2nd-last data
-            blockwise_gemm.Run(
-                c_m10_m11_n10_n11_thread_desc, a_block_even_buf, b_block_even_buf, c_thread_buf);
-
-            // LDS double buffer: store last data to LDS
-            a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_odd_buf);
-            b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_odd_buf);
-
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_m10_m11_n10_n11_thread_desc, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-        }
-        else // if has 1 iteration left
-        {
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_m10_m11_n10_n11_thread_desc, a_block_even_buf, b_block_even_buf, c_thread_buf);
-        }
-
-        // output: register to global memory
-        {
-            constexpr auto c_m0_m10_m11_n0_n10_n11_thread_desc =
-                make_naive_tensor_descriptor_packed(
-                    make_tuple(I1,
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I0]>{},
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I1]>{},
-                               I1,
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I2]>{},
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I3]>{}));
-
-            const auto c_m10_m11_n10_n11_thread_origin_idx_on_block =
-                blockwise_gemm.CalculateCM0M1N0N1ThreadOriginOnBlock(get_thread_local_1d_id());
-
-            ThreadwiseTensorSliceTransfer_v1r3<
-                FloatAcc,
-                FloatC,
-                decltype(c_m0_m10_m11_n0_n10_n11_thread_desc),
-                decltype(c_m0_m10_m11_n0_n10_n11_grid_desc),
-                Sequence<1,
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I0],
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I1],
-                         1,
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I2],
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I3]>,
-                CThreadTransferSrcDstAccessOrder,
-                CThreadTransferSrcDstVectorDim,
-                CThreadTransferDstScalarPerVector,
-                CGlobalMemoryDataOperation,
-                1,
-                true>{c_m0_m10_m11_n0_n10_n11_grid_desc,
-                      make_multi_index(im0,
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I0],
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I1],
-                                       in0,
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I2],
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I3])}
-                .Run(c_m0_m10_m11_n0_n10_n11_thread_desc,
-                     make_tuple(I0, I0, I0, I0, I0, I0),
-                     c_thread_buf,
-                     c_m0_m10_m11_n0_n10_n11_grid_desc,
-                     c_grid_buf,
-                     CGridStepHacks{});
-        }
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v2.hpp
deleted file mode 100644
index b1dfb0c73fc92a19662e01df31090008cdef87ff..0000000000000000000000000000000000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v2.hpp
+++ /dev/null
@@ -1,461 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_GEMM_V2_HPP
-#define CK_GRIDWISE_GEMM_V2_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "blockwise_gemm_dlops_v3.hpp"
-
-namespace ck {
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGlobalDesc,
-          typename BGlobalDesc,
-          typename CGlobalDesc,
-          index_t KPerBlock,
-          index_t HoPerBlock,
-          index_t WoPerBlock,
-          index_t EPerBlock,
-          index_t KPerThread,
-          index_t HoPerThread,
-          index_t WoPerThread,
-          index_t EPerThread,
-          typename ABlockTransferThreadSliceLengths_E_K,
-          typename ABlockTransferThreadClusterLengths_E_K,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_K,
-          bool AThreadTransferSrcResetCoordinateAfterRun,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          bool BThreadTransferSrcResetCoordinateAfterRun,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGlobalStepHacks,
-          typename BGlobalStepHacks,
-          typename CGlobalStepHacks,
-          typename AGlobalMoveSliceWindowStepHacks,
-          typename BGlobalMoveSliceWindowStepHacks>
-struct GridwiseGemmDlops_km_kn_mn_v3
-{
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        constexpr auto E = EPerBlock * 3 * 3;
-
-        constexpr auto max_lds_align =
-            math::lcm(Number<ABlockTransferDstScalarPerVector_K>{}, Number<KPerBlock>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_e_k_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E>{}, Number<KPerBlock>{}), max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_space_size =
-            math::integer_least_multiple(a_e_k_desc.GetElementSpaceSize(), max_lds_align);
-
-        return a_block_space_size * sizeof(FloatAB);
-    }
-
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const AGlobalDesc& a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const BGlobalDesc& b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const CGlobalDesc& c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        FloatAB* __restrict__ p_shared_block,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-        constexpr auto I2 = Number<2>{};
-        constexpr auto I3 = Number<3>{};
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e_k_global_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e_n_ho_wo_global_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k_n_ho_wo_global_desc.GetElementSpaceSize());
-
-        constexpr auto E = EPerBlock * 3 * 3;
-
-        // const auto E = a_e_k_global_desc.GetLength(I0);
-        const auto K = a_e_k_global_desc.GetLength(I1);
-
-        const auto N  = b_e_n_ho_wo_global_desc.GetLength(I1);
-        const auto Ho = b_e_n_ho_wo_global_desc.GetLength(I2);
-        const auto Wo = b_e_n_ho_wo_global_desc.GetLength(I3);
-
-// divide block work by [M, N]
-#if 0
-        const auto ho_block_work_num  = Ho / Number<HoPerBlock>{};
-        const auto wo_block_work_num  = Wo / Number<WoPerBlock>{};
-        const auto hwo_block_work_num = ho_block_work_num * wo_block_work_num;
-
-        const index_t k_block_work_id   = get_block_1d_id() / hwo_block_work_num;
-        const index_t hwo_block_work_id = get_block_1d_id() - k_block_work_id * hwo_block_work_num;
-
-        const index_t ho_block_work_id = hwo_block_work_id / wo_block_work_num;
-        const index_t wo_block_work_id = hwo_block_work_id - ho_block_work_id * wo_block_work_num;
-#else
-        // Hack: this force result into SGPR
-        const index_t ho_block_work_num  = __builtin_amdgcn_readfirstlane(Ho / HoPerBlock);
-        const index_t wo_block_work_num  = __builtin_amdgcn_readfirstlane(Wo / WoPerBlock);
-        const index_t hwo_block_work_num = ho_block_work_num * wo_block_work_num;
-
-        const index_t k_block_work_id =
-            __builtin_amdgcn_readfirstlane(get_block_1d_id() / hwo_block_work_num);
-        const index_t hwo_block_work_id = get_block_1d_id() - k_block_work_id * hwo_block_work_num;
-
-        const index_t ho_block_work_id =
-            __builtin_amdgcn_readfirstlane(hwo_block_work_id / wo_block_work_num);
-        const index_t wo_block_work_id = hwo_block_work_id - ho_block_work_id * wo_block_work_num;
-#endif
-
-        // lds max alignment
-        constexpr auto max_lds_align =
-            math::lcm(Number<ABlockTransferDstScalarPerVector_K>{}, Number<KPerBlock>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_e_k_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<EPerBlock>{}, Number<KPerBlock>{}), max_lds_align);
-
-        constexpr auto a_e_k_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E>{}, Number<KPerBlock>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_e_n_ho_wo_block_desc = make_naive_tensor_descriptor_packed(make_tuple(
-            Number<EPerBlock>{}, Number<1>{}, Number<HoPerBlock>{}, Number<WoPerBlock>{}));
-
-        // c_thread_mtx definition: this is a mess
-        // TODO:: more elegent way of defining c_thread_mtx
-        constexpr auto c_k_n_ho_wo_thread_desc = make_naive_tensor_descriptor_packed(make_tuple(
-            Number<KPerThread>{}, Number<1>{}, Number<HoPerThread>{}, Number<WoPerThread>{}));
-
-        auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v3<BlockSize,
-                                                 FloatAB,
-                                                 FloatAB,
-                                                 FloatAcc,
-                                                 decltype(a_e_k_block_desc),
-                                                 decltype(b_e_n_ho_wo_block_desc),
-                                                 decltype(c_k_n_ho_wo_thread_desc),
-                                                 KPerThread,
-                                                 HoPerThread,
-                                                 WoPerThread,
-                                                 EPerThread,
-                                                 ABlockTransferSrcScalarPerVector,
-                                                 ABlockTransferDstScalarPerVector_K>{};
-
-        auto c_thread_mtx_index = blockwise_gemm.GetBeginOfThreadMatrixC(get_thread_local_1d_id());
-
-        const auto k_thread_id  = c_thread_mtx_index.k;
-        const auto ho_thread_id = c_thread_mtx_index.h;
-        const auto wo_thread_id = c_thread_mtx_index.w;
-
-        const index_t k_block_data_on_global  = k_block_work_id * KPerBlock;
-        const index_t ho_block_data_on_global = ho_block_work_id * HoPerBlock;
-        const index_t wo_block_data_on_global = wo_block_work_id * WoPerBlock;
-
-        const index_t ho_thread_data_on_global =
-            ho_block_data_on_global + ho_thread_id * HoPerThread;
-        const index_t wo_thread_data_on_global =
-            wo_block_data_on_global + wo_thread_id * WoPerThread;
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<E, KPerBlock>,
-                                            ABlockTransferThreadSliceLengths_E_K,
-                                            ABlockTransferThreadClusterLengths_E_K,
-                                            ABlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(a_e_k_global_desc),
-                                            decltype(a_e_k_desc),
-                                            ABlockTransferSrcAccessOrder,
-                                            Sequence<0, 1>,
-                                            ABlockTransferSrcVectorDim,
-                                            1,
-                                            ABlockTransferSrcScalarPerVector,
-                                            ABlockTransferDstScalarPerVector_K,
-                                            1,
-                                            1,
-                                            AThreadTransferSrcResetCoordinateAfterRun,
-                                            true>(a_e_k_global_desc,
-                                                  make_multi_index(0, k_block_data_on_global),
-                                                  a_e_k_desc,
-                                                  make_multi_index(0, 0));
-
-        constexpr auto b_e_n_ho_wo_thread_desc = make_naive_tensor_descriptor_packed(make_tuple(
-            Number<EPerBlock>{}, Number<1>{}, Number<HoPerThread>{}, Number<WoPerThread>{}));
-
-        auto b_threadwise_transfer =
-            ThreadwiseTensorSliceTransfer_v2<FloatAB,
-                                             FloatAB,
-                                             decltype(b_e_n_ho_wo_global_desc),
-                                             decltype(b_e_n_ho_wo_thread_desc),
-                                             Sequence<EPerBlock, 1, HoPerThread, WoPerThread>,
-                                             BBlockTransferSrcAccessOrder,
-                                             BBlockTransferSrcVectorDim,
-                                             BBlockTransferSrcScalarPerVector,
-                                             1,
-                                             true>(
-                b_e_n_ho_wo_global_desc,
-                make_multi_index(0, 0, ho_thread_data_on_global, wo_thread_data_on_global));
-
-        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_shared_block, a_e_k_desc.GetElementSpaceSize());
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k_n_ho_wo_thread_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        // initialize output thread tensor
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_k_n_ho_wo_thread_desc),
-                                    Sequence<KPerThread, 1, HoPerThread, WoPerThread>>{}
-            .Run(c_k_n_ho_wo_thread_desc, make_tuple(I0, I0, I0, I0), c_thread_buf, FloatAcc{0});
-
-        constexpr auto b_thread_slice_copy_step = make_multi_index(EPerBlock, 0, 0, 0);
-
-        // hack to control index calculation when iterating over A and B matrix for threadwise copy
-        constexpr auto a_e_k_global_step_hacks       = AGlobalStepHacks{};
-        constexpr auto b_e_n_ho_wo_global_step_hacks = BGlobalStepHacks{};
-
-        // hack to control index calculation when move slice window for A and B matrix for
-        // threadwise copy
-        constexpr auto a_e_k_global_move_slice_window_step_hack = AGlobalMoveSliceWindowStepHacks{};
-        constexpr auto b_e_n_ho_wo_global_move_slice_window_step_hack =
-            BGlobalMoveSliceWindowStepHacks{};
-
-        // double regsiter buffer for b
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAB,
-                     b_e_n_ho_wo_thread_desc.GetElementSpaceSize(),
-                     true>
-            b_thread_even_buf, b_thread_odd_buf;
-
-        // LDS double buffer: preload data
-        {
-            a_blockwise_copy.RunRead(a_e_k_global_desc, a_global_buf, a_e_k_global_step_hacks);
-
-            b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                      b_global_buf,
-                                      b_e_n_ho_wo_thread_desc,
-                                      make_tuple(I0, I0, I0, I0),
-                                      b_thread_even_buf,
-                                      b_e_n_ho_wo_global_step_hacks);
-
-            a_blockwise_copy.RunWrite(a_e_k_desc, a_block_buf);
-        }
-
-        __syncthreads();
-
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t e_block_data_begin = 0;
-
-            // LDS double buffer: main body
-            // use Do-While loop instead of For loop to simplify control flow
-            do
-            {
-                // even iteration
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e_n_ho_wo_global_desc,
-                                                         b_thread_slice_copy_step);
-
-                b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                          b_global_buf,
-                                          b_e_n_ho_wo_thread_desc,
-                                          make_tuple(I0, I0, I0, I0),
-                                          b_thread_odd_buf,
-                                          b_e_n_ho_wo_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                // TODO: @Zhang Jing: blockwise gemm should be able to move slice window
-                blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                blockwise_gemm.MoveASliceWindow(a_e_k_block_desc, make_tuple(EPerBlock, 0));
-
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e_n_ho_wo_global_desc,
-                                                         b_thread_slice_copy_step);
-
-                b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                          b_global_buf,
-                                          b_e_n_ho_wo_thread_desc,
-                                          make_tuple(I0, I0, I0, I0),
-                                          b_thread_even_buf,
-                                          b_e_n_ho_wo_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-
-                blockwise_gemm.MoveASliceWindow(a_e_k_block_desc, make_tuple(EPerBlock, 0));
-
-                e_block_data_begin += 2 * EPerBlock;
-
-            } while(e_block_data_begin < E - 2 * EPerBlock);
-        }
-
-        // LDS double buffer: tail
-        if constexpr(HasDoubleTailKBlockLoop) // if has 2 iteration left
-        {
-            b_threadwise_transfer.MoveSrcSliceWindow(b_e_n_ho_wo_global_desc,
-                                                     b_thread_slice_copy_step);
-
-            b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                      b_global_buf,
-                                      b_e_n_ho_wo_thread_desc,
-                                      make_tuple(I0, I0, I0, I0),
-                                      b_thread_odd_buf,
-                                      b_e_n_ho_wo_global_step_hacks);
-
-            // LDS double buffer: GEMM on 2nd-last data
-            blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-            blockwise_gemm.MoveASliceWindow(a_e_k_block_desc, make_tuple(EPerBlock, 0));
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-        }
-        else // if has 1 iteration left
-        {
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-        }
-
-        // output: register to global memory
-        {
-            // hack to control index calculation when iterating over c_k_n_ho_wo_global tensor
-            constexpr auto c_k_n_ho_wo_global_tensor_step_hacks = CGlobalStepHacks{};
-
-            const index_t k_thread_data_on_global =
-                k_block_data_on_global + k_thread_id * KPerThread;
-
-            ThreadwiseTensorSliceTransfer_v1r3<FloatAcc,
-                                               FloatC,
-                                               decltype(c_k_n_ho_wo_thread_desc),
-                                               decltype(c_k_n_ho_wo_global_desc),
-                                               Sequence<KPerThread, 1, HoPerThread, WoPerThread>,
-                                               CThreadTransferSrcDstAccessOrder,
-                                               CThreadTransferSrcDstVectorDim,
-                                               CThreadTransferDstScalarPerVector,
-                                               CGlobalMemoryDataOperation,
-                                               1,
-                                               true>(
-                c_k_n_ho_wo_global_desc,
-                make_multi_index(
-                    k_thread_data_on_global, 0, ho_thread_data_on_global, wo_thread_data_on_global))
-                .Run(c_k_n_ho_wo_thread_desc,
-                     make_tuple(I0, I0, I0, I0),
-                     c_thread_buf,
-                     c_k_n_ho_wo_global_desc,
-                     c_global_buf,
-                     c_k_n_ho_wo_global_tensor_step_hacks);
-        }
-    }
-
-    // pass tensor descriptor by reference
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const AGlobalDesc& a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const BGlobalDesc& b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const CGlobalDesc& c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        constexpr index_t shared_block_size = GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-        __shared__ FloatAB p_shared_block[shared_block_size];
-
-        Run(a_e_k_global_desc,
-            p_a_global,
-            b_e_n_ho_wo_global_desc,
-            p_b_global,
-            c_k_n_ho_wo_global_desc,
-            p_c_global,
-            p_shared_block,
-            integral_constant<bool, HasMainKBlockLoop>{},
-            integral_constant<bool, HasDoubleTailKBlockLoop>{});
-    }
-
-    // pass tensor descriptors by their pointers
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const AGlobalDesc* p_a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const BGlobalDesc* p_b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const CGlobalDesc* p_c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        const auto a_e_k_global_desc       = *p_a_e_k_global_desc;
-        const auto b_e_n_ho_wo_global_desc = *p_b_e_n_ho_wo_global_desc;
-        const auto c_k_n_ho_wo_global_desc = *p_c_k_n_ho_wo_global_desc;
-
-        Run(a_e_k_global_desc,
-            p_a_global,
-            b_e_n_ho_wo_global_desc,
-            p_b_global,
-            c_k_n_ho_wo_global_desc,
-            p_c_global,
-            integral_constant<bool, HasMainKBlockLoop>{},
-            integral_constant<bool, HasDoubleTailKBlockLoop>{});
-    }
-
-    // pass tensor descriptors by void*
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const void* p_a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const void* p_b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const void* p_c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        const auto a_e_k_global_desc = *reinterpret_cast<const AGlobalDesc*>(p_a_e_k_global_desc);
-        const auto b_e_n_ho_wo_global_desc =
-            *reinterpret_cast<const BGlobalDesc*>(p_b_e_n_ho_wo_global_desc);
-        const auto c_k_n_ho_wo_global_desc =
-            *reinterpret_cast<const CGlobalDesc*>(p_c_k_n_ho_wo_global_desc);
-
-        Run(a_e_k_global_desc,
-            p_a_global,
-            b_e_n_ho_wo_global_desc,
-            p_b_global,
-            c_k_n_ho_wo_global_desc,
-            p_c_global,
-            integral_constant<bool, HasMainKBlockLoop>{},
-            integral_constant<bool, HasDoubleTailKBlockLoop>{});
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v3.hpp
deleted file mode 100644
index ace844338414668030e5dd6aeeaf559e1a7c1060..0000000000000000000000000000000000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v3.hpp
+++ /dev/null
@@ -1,1597 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_GEMM_V3_HPP
-#define CK_GRIDWISE_GEMM_V3_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_set.hpp"
-#include "blockwise_gemm_dlops_v3.hpp"
-
-namespace ck {
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_E0_E1_K0_K1_E2,
-          typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-          typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-          typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-          bool HasMainE0BlockLoop,
-          ActivTypeEnum ActivType>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v3(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            const FloatC* __restrict__ p_bias_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AGridDesc_E0_E1_K0_K1_E2 a_e0_e1_k0_k1_e2_grid_desc,
-            const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-            const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-            const CBlockIdToBlockClusterAdaptor_K_N_H_W cblockid_to_k_n_h_w_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::ConvBiasActiv(p_a_grid,
-                                p_b_grid,
-                                p_bias_grid,
-                                p_c_grid,
-                                p_shared_block,
-                                a_e0_e1_k0_k1_e2_grid_desc,
-                                b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                                cblockid_to_k_n_h_w_block_cluster_adaptor,
-                                integral_constant<bool, HasMainE0BlockLoop>{},
-                                integral_constant<ActivTypeEnum, ActivType>{});
-}
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_E0_E1_K0_K1_E2,
-          typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-          typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-          typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-          typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-          bool HasMainE0BlockLoop,
-          ActivTypeEnum ActivType>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v3_resize_add(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            const FloatC* __restrict__ p_bias_grid,
-            FloatC* __restrict__ p_d_grid,
-            const AGridDesc_E0_E1_K0_K1_E2 a_e0_e1_k0_k1_e2_grid_desc,
-            const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-            const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-            const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-            const CBlockIdToBlockClusterAdaptor_K_N_H_W cblockid_to_k_n_h_w_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::ConvBiasActivResizeAdd(p_a_grid,
-                                         p_b_grid,
-                                         p_bias_grid,
-                                         p_d_grid,
-                                         p_shared_block,
-                                         a_e0_e1_k0_k1_e2_grid_desc,
-                                         b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                         c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                                         d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                                         cblockid_to_k_n_h_w_block_cluster_adaptor,
-                                         integral_constant<bool, HasMainE0BlockLoop>{},
-                                         integral_constant<ActivTypeEnum, ActivType>{});
-}
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_E0_E1_K0_K1_E2,
-          typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-          typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-          typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-          typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-          bool HasMainE0BlockLoop,
-          ActivTypeEnum ActivType>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v3_maxpool(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            const FloatC* __restrict__ p_bias_grid,
-            FloatC* __restrict__ p_c_grid,
-            FloatC* __restrict__ p_d_grid,
-            const AGridDesc_E0_E1_K0_K1_E2 a_e0_e1_k0_k1_e2_grid_desc,
-            const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-            const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-            const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-            const CBlockIdToBlockClusterAdaptor_K_N_H_W cblockid_to_k_n_h_w_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::ConvBiasActivMaxpool(p_a_grid,
-                                       p_b_grid,
-                                       p_bias_grid,
-                                       p_c_grid,
-                                       p_d_grid,
-                                       p_shared_block,
-                                       a_e0_e1_k0_k1_e2_grid_desc,
-                                       b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                       c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                                       d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                                       cblockid_to_k_n_h_w_block_cluster_adaptor,
-                                       integral_constant<bool, HasMainE0BlockLoop>{},
-                                       integral_constant<ActivTypeEnum, ActivType>{});
-}
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_E0_E1_K_E2,
-          typename BGridDesc_E0_E1_N_Ho_Wo_E2,
-          typename CGridDesc_K_N_Ho_Wo,
-          typename DGridDesc_K_N_Hx_Wx,
-          index_t E1_,
-          index_t E2_,
-          index_t K2_,
-          index_t KPerBlock,
-          index_t HoPerBlock,
-          index_t WoPerBlock,
-          index_t E1PerBlock,
-          index_t KPerThread,
-          index_t HoPerThread,
-          index_t WoPerThread,
-          index_t EPerThread,
-          typename ABlockTransferThreadSliceLengths_E0_E1_K0_K1_E2,
-          typename ABlockTransferThreadClusterLengths_E0_E1_K0_K1_E2,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_E2,
-          bool AThreadTransferSrcResetCoordinateAfterRun,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          bool BThreadTransferSrcResetCoordinateAfterRun,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGlobalStepHacks,
-          typename BGlobalStepHacks,
-          typename CGlobalStepHacks,
-          typename DGlobalStepHacks,
-          typename AGlobalMoveSliceWindowStepHacks,
-          typename BGlobalMoveSliceWindowStepHacks>
-struct GridwiseGemmDlops_km_kn_mn_v3
-{
-
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-    static constexpr auto I4 = Number<4>{};
-    static constexpr auto I5 = Number<5>{};
-
-    static constexpr auto E1 = Number<E1_>{};
-    static constexpr auto E2 = Number<E2_>{};
-    static constexpr auto K2 = Number<K2_>{};
-
-    static constexpr auto NPerBlock = I1;
-
-    static constexpr FloatAcc alpha = 0.3;
-
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        constexpr auto max_lds_align = Number<ABlockTransferDstScalarPerVector_E2>{};
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_e0_e1_k1_e2_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(I1, Number<E1>{}, Number<KPerBlock>{}, Number<E2>{}), max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_space_size = math::integer_least_multiple(
-            a_e0_e1_k1_e2_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        return a_block_space_size * sizeof(FloatAB);
-    }
-
-    __host__ __device__ static constexpr index_t
-    CalculateGridSize(const CGridDesc_K_N_Ho_Wo& c_k_n_ho_wo_grid_desc)
-    {
-        const auto K  = c_k_n_ho_wo_grid_desc.GetLength(I0);
-        const auto N  = c_k_n_ho_wo_grid_desc.GetLength(I1);
-        const auto Ho = c_k_n_ho_wo_grid_desc.GetLength(I2);
-        const auto Wo = c_k_n_ho_wo_grid_desc.GetLength(I3);
-
-        const auto K0 = K / KPerBlock;
-        const auto N0 = N / NPerBlock;
-        const auto H0 = Ho / HoPerBlock;
-        const auto W0 = Wo / WoPerBlock;
-
-        const index_t grid_size = K0 * N0 * H0 * W0;
-
-        return grid_size;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainE0BlockLoop(const index_t E0)
-    {
-        const bool has_main_e0_block_loop = E0 > 1;
-
-        return has_main_e0_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainE1BlockLoop()
-    {
-        const bool has_main_e1_block_loop = ((E1 + E1PerBlock) / (2 * E1PerBlock)) > 1;
-
-        return has_main_e1_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasDoubleTailE1BlockLoop()
-    {
-        const bool has_double_tail_e1_block_loop = (E1 / E1PerBlock) % 2 == 0;
-
-        return has_double_tail_e1_block_loop;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeAE0E1K0K1E2GridDescriptor(const AGridDesc_E0_E1_K_E2& a_e0_e1_k_e2_grid_desc)
-    {
-        const auto E0 = a_e0_e1_k_e2_grid_desc.GetLength(I0);
-        const auto K  = a_e0_e1_k_e2_grid_desc.GetLength(I2);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-        const auto a_e0_e1_k0_k1_e2_grid_desc = transform_tensor_descriptor(
-            a_e0_e1_k_e2_grid_desc,
-            make_tuple(make_pass_through_transform(E0),
-                       make_pass_through_transform(E1),
-                       make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(E2)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-        return a_e0_e1_k0_k1_e2_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto MakeBE0E1NH0H1H2W0W1W2E2GridDescriptor(
-        const BGridDesc_E0_E1_N_Ho_Wo_E2& b_e0_e1_n_ho_wo_e2_grid_desc)
-    {
-        const auto E0 = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I0);
-        // const auto E1 = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I1);
-        const auto N  = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I2);
-        const auto Ho = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I3);
-        const auto Wo = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I4);
-        // const auto E2 = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I5);
-
-        const auto H2 = Number<HoPerThread>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-        const auto H0 = Ho / (H1 * H2);
-
-        const auto W2 = Number<WoPerThread>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-        const auto W0 = Wo / (W1 * W2);
-
-        const auto b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc =
-            transform_tensor_descriptor(b_e0_e1_n_ho_wo_e2_grid_desc,
-                                        make_tuple(make_pass_through_transform(E0),
-                                                   make_pass_through_transform(E1),
-                                                   make_pass_through_transform(N),
-                                                   make_unmerge_transform(make_tuple(H0, H1, H2)),
-                                                   make_unmerge_transform(make_tuple(W0, W1, W2)),
-                                                   make_pass_through_transform(E2)),
-                                        make_tuple(Sequence<0>{},
-                                                   Sequence<1>{},
-                                                   Sequence<2>{},
-                                                   Sequence<3>{},
-                                                   Sequence<4>{},
-                                                   Sequence<5>{}),
-                                        make_tuple(Sequence<0>{},
-                                                   Sequence<1>{},
-                                                   Sequence<2>{},
-                                                   Sequence<3, 4, 5>{},
-                                                   Sequence<6, 7, 8>{},
-                                                   Sequence<9>{}));
-
-        return b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCK0K1NH0H1H2W0W1W2GridDescriptor(const CGridDesc_K_N_Ho_Wo& c_k_n_ho_wo_grid_desc)
-    {
-        const auto K  = c_k_n_ho_wo_grid_desc.GetLength(I0);
-        const auto N  = c_k_n_ho_wo_grid_desc.GetLength(I1);
-        const auto Ho = c_k_n_ho_wo_grid_desc.GetLength(I2);
-        const auto Wo = c_k_n_ho_wo_grid_desc.GetLength(I3);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-        const auto H2 = Number<HoPerThread>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-        const auto H0 = Ho / (H1 * H2);
-
-        const auto W2 = Number<WoPerThread>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-        const auto W0 = Wo / (W1 * W2);
-
-        const auto c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc = transform_tensor_descriptor(
-            c_k_n_ho_wo_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(N),
-                       make_unmerge_transform(make_tuple(H0, H1, H2)),
-                       make_unmerge_transform(make_tuple(W0, W1, W2))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0, 1>{}, Sequence<2>{}, Sequence<3, 4, 5>{}, Sequence<6, 7, 8>{}));
-
-        return c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeDK0K1NH0H1HxW0W1WxGridDescriptorMaxPool(const DGridDesc_K_N_Hx_Wx& d_k_n_hx_wx_grid_desc)
-    {
-        const auto K  = d_k_n_hx_wx_grid_desc.GetLength(I0);
-        const auto N  = d_k_n_hx_wx_grid_desc.GetLength(I1);
-        const auto Hx = d_k_n_hx_wx_grid_desc.GetLength(I2);
-        const auto Wx = d_k_n_hx_wx_grid_desc.GetLength(I3);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-#if CK_EXPERIMENTAL_STATIC_TENSOR_DESCRIPTOR
-        const auto H2 = Number<HoPerThread / 2>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-        const auto H0 = Number<Hx / (H1 * H2)>{};
-
-        const auto W2 = Number<WoPerThread / 2>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-        const auto W0 = Number<Wx / (W1 * W2)>{};
-#else
-        const auto H2 = HoPerThread / 2;
-        const auto H1 = HoPerBlock / HoPerThread;
-        const auto H0 = Hx / (H1 * H2);
-
-        const auto W2 = WoPerThread / 2;
-        const auto W1 = WoPerBlock / WoPerThread;
-        const auto W0 = Wx / (W1 * W2);
-#endif
-
-        const auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc = transform_tensor_descriptor(
-            d_k_n_hx_wx_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(N),
-                       make_unmerge_transform(make_tuple(H0, H1, H2)),
-                       make_unmerge_transform(make_tuple(W0, W1, W2))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0, 1>{}, Sequence<2>{}, Sequence<3, 4, 5>{}, Sequence<6, 7, 8>{}));
-
-        return d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeDK0K1NH0H1HxW0W1WxGridDescriptorResizeAdd(const DGridDesc_K_N_Hx_Wx& d_k_n_hx_wx_grid_desc)
-    {
-        const auto K  = d_k_n_hx_wx_grid_desc.GetLength(I0);
-        const auto N  = d_k_n_hx_wx_grid_desc.GetLength(I1);
-        const auto Hx = d_k_n_hx_wx_grid_desc.GetLength(I2);
-        const auto Wx = d_k_n_hx_wx_grid_desc.GetLength(I3);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-        const auto H2 = Number<HoPerThread * 2>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-
-        const auto W2 = Number<WoPerThread * 2>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-
-#if CK_EXPERIMENTAL_STATIC_TENSOR_DESCRIPTOR
-        const auto H0 = Number<Hx / (H1 * H2)>{};
-        const auto W0 = Number<Wx / (W1 * W2)>{};
-#else
-        const auto H0 = Hx / (H1 * H2);
-        const auto W0 = Wx / (W1 * W2);
-#endif
-
-        const auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc = transform_tensor_descriptor(
-            d_k_n_hx_wx_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(N),
-                       make_unmerge_transform(make_tuple(H0, H1, H2)),
-                       make_unmerge_transform(make_tuple(W0, W1, W2))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0, 1>{}, Sequence<2>{}, Sequence<3, 4, 5>{}, Sequence<6, 7, 8>{}));
-
-        return d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCBlockIdToKNHoWoBlockClusterAdaptor(const CGridDesc_K_N_Ho_Wo& c_k_n_ho_wo_grid_desc)
-    {
-        const auto K  = c_k_n_ho_wo_grid_desc.GetLength(I0);
-        const auto N  = c_k_n_ho_wo_grid_desc.GetLength(I1);
-        const auto Ho = c_k_n_ho_wo_grid_desc.GetLength(I2);
-        const auto Wo = c_k_n_ho_wo_grid_desc.GetLength(I3);
-
-#if CK_EXPERIMENTAL_STATIC_TENSOR_DESCRIPTOR
-        const auto K0 = Number<K / KPerBlock>{};
-        const auto N0 = Number<N / NPerBlock>{};
-        const auto H0 = Number<Ho / HoPerBlock>{};
-        const auto W0 = Number<Wo / WoPerBlock>{};
-#else
-        const auto K0 = K / KPerBlock;
-        const auto N0 = N / NPerBlock;
-        const auto H0 = Ho / HoPerBlock;
-        const auto W0 = Wo / WoPerBlock;
-#endif
-
-        const auto cblockid_to_k_n_ho_wo_block_cluster_adaptor = make_single_stage_tensor_adaptor(
-            make_tuple(make_merge_transform(make_tuple(K0, N0, H0, W0))),
-            make_tuple(Sequence<0, 1, 2, 3>{}),
-            make_tuple(Sequence<0>{}));
-
-        return cblockid_to_k_n_ho_wo_block_cluster_adaptor;
-    }
-
-    // using AGridDesc_E0_E1_K0_K1_E2 =
-    // decltype(MakeAE0E1K0K1E2GridDescriptor(AGridDesc_E0_E1_K_E2{}));
-    // using BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 =
-    // decltype(MakeBE0E1NH0H1H2W0W1W2E2GridDescriptor(BGridDesc_E0_E1_N_Ho_Wo_E2{}));
-    // using CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 =
-    // decltype(MakeCK0K1NH0H1H2W0W1W2GridDescriptor(CGridDesc_K_N_Ho_Wo{}));
-    // using DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx =
-    // decltype(MakeDK0K1NH0H1HxW0W1WxGridDescriptor(DGridDesc_K_N_Hx_Wx{}));
-
-    using CBlockIdToBlockClusterAdaptor_K_N_H_W =
-        decltype(MakeCBlockIdToKNHoWoBlockClusterAdaptor(CGridDesc_K_N_Ho_Wo{}));
-
-    template <typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2>
-    __host__ __device__ static constexpr auto MakeBiasK0K1GridDescriptor(
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc)
-    {
-        const auto K0 = c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetLength(I0);
-        const auto K1 = c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetLength(I1);
-
-        return make_naive_tensor_descriptor_packed(make_tuple(K0, K1));
-    }
-
-    __host__ __device__ static constexpr auto MakeCK1NH2W2ThreadDescriptor()
-    {
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<KPerThread>{}, I1, Number<HoPerThread>{}, Number<WoPerThread>{}));
-        return c_k1_n_h2_w2_thread_gemm_desc;
-    }
-
-    // using CThreadDesc_K1_N_H2_W2 = decltype(MakeCK1NH2W2ThreadDescriptor());
-
-    __host__ __device__ static constexpr auto GetBlockWiseGemm()
-    {
-        constexpr auto max_lds_align = Number<ABlockTransferDstScalarPerVector_E2>{};
-
-        constexpr auto a_e1_k1_e2_block_gemm_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E1PerBlock>{}, Number<KPerBlock>{}, Number<E2>{}), max_lds_align);
-
-        constexpr auto b_e1_n_h_w_e2_block_gemm_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(Number<E1PerBlock>{},
-                                                           I1,
-                                                           Number<HoPerBlock>{},
-                                                           Number<WoPerBlock>{},
-                                                           Number<E2>{}));
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v3<BlockSize,
-                                                 FloatAB,
-                                                 FloatAB,
-                                                 FloatAcc,
-                                                 decltype(a_e1_k1_e2_block_gemm_desc),
-                                                 decltype(b_e1_n_h_w_e2_block_gemm_desc),
-                                                 decltype(c_k1_n_h2_w2_thread_gemm_desc),
-                                                 EPerThread,
-                                                 K2>{};
-
-        return blockwise_gemm;
-    }
-
-    __device__ static constexpr auto GetCThreadIndex()
-    {
-        auto blockwise_gemm = GetBlockWiseGemm();
-        auto c_thread_mtx_index =
-            blockwise_gemm.GetBeginOfCThreadDesc_K_N_Ho_Wo(get_thread_local_1d_id());
-
-        return c_thread_mtx_index;
-    };
-
-    __device__ static constexpr auto GetCBlockIndex(
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor)
-    {
-        const auto c_k_n_h_w_block_cluster_idx =
-            cblockid_to_k_n_h_w_block_cluster_adaptor.CalculateBottomIndex(
-                make_multi_index(get_block_1d_id()));
-        return c_k_n_h_w_block_cluster_idx;
-    }
-
-    template <typename BiasGlobalBuff,
-              typename CThreadBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename BiasGridDesc_K0_K1,
-              typename CThreadDesc_K1_N_H2_W2>
-    __device__ static void BiasOp(BiasGlobalBuff& bias_global_buf,
-                                  CThreadBuff& c_thread_buf,
-                                  const CBlockIndex& c_block_idx,
-                                  const CThreadIndex& c_thread_idx,
-                                  const BiasGridDesc_K0_K1& bias_k0_k1_grid_desc,
-                                  const CThreadDesc_K1_N_H2_W2&)
-
-    {
-        const index_t k_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-
-        const auto k_thread_id = c_thread_idx[I0];
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        constexpr auto bias_k0_k1_thread_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1, Number<KPerThread>{}));
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatC,
-                     bias_k0_k1_thread_desc.GetElementSpaceSize(),
-                     true>
-            bias_thread_buf;
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        auto bias_threadwise_transfer =
-            ThreadwiseTensorSliceTransfer_v2<FloatC,
-                                             FloatC,
-                                             decltype(bias_k0_k1_grid_desc),
-                                             decltype(bias_k0_k1_thread_desc),
-                                             Sequence<I1, Number<KPerThread>{}>,
-                                             Sequence<0, 1>,
-                                             1,
-                                             CThreadTransferDstScalarPerVector,
-                                             false,
-                                             true>(
-                bias_k0_k1_grid_desc, make_multi_index(k_block_work_id, k_thread_data_on_global));
-
-        constexpr auto bias_k0_k1_global_tensor_step_hacks = make_tuple(
-            make_tuple(Sequence<0>{}, Sequence<0>{}), make_tuple(Sequence<0>{}, Sequence<0>{}));
-
-        bias_threadwise_transfer.Run(bias_k0_k1_grid_desc,
-                                     bias_global_buf,
-                                     bias_k0_k1_thread_desc,
-                                     make_tuple(I0, I0),
-                                     bias_thread_buf,
-                                     bias_k0_k1_global_tensor_step_hacks);
-
-        static_for<0, KPerThread, 1>{}([&](auto ki) {
-            static_for<0, HoPerThread, 1>{}([&](auto hi) {
-                static_for<0, WoPerThread, 1>{}([&](auto wi) {
-                    constexpr index_t c_offset =
-                        c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(make_tuple(ki, 0, hi, wi));
-                    c_thread_buf(Number<c_offset>{}) =
-                        c_thread_buf[Number<c_offset>{}] + bias_thread_buf[ki];
-                });
-            });
-        });
-    }
-
-    template <typename CThreadBuff, typename CThreadDesc_K1_N_H2_W2, ActivTypeEnum activ_type_>
-    __device__ static void Activation(CThreadBuff& c_thread_buf,
-                                      const CThreadDesc_K1_N_H2_W2&,
-                                      integral_constant<ActivTypeEnum, activ_type_>)
-    {
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        static_for<0, c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(), 1>{}([&](auto i) {
-            if constexpr(activ_type_ == 1)
-            {
-                c_thread_buf(i) = c_thread_buf[i] >= 0 ? c_thread_buf[i] : alpha * c_thread_buf[i];
-            }
-            else if constexpr(activ_type_ == 2)
-            {
-                FloatAcc x = 1.0 + exp(-c_thread_buf[i]);
-
-                asm volatile("\n \
-                        v_rcp_f32 %0, %1 \n"
-                             : "=v"(x)
-                             : "0"(x));
-
-                c_thread_buf(i) = x;
-            }
-        });
-    }
-
-    template <typename CThreadBuff,
-              typename CGlobalBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2>
-    __device__ static void
-    WriteOut(const CThreadBuff& c_thread_buf,
-             CGlobalBuff& c_global_buf,
-             const CBlockIndex& c_block_idx,
-             const CThreadIndex& c_thread_idx,
-             const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc)
-    {
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        const auto k_thread_id  = c_thread_idx[I0];
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        // hack to control index calculation when iterating over c_k_n_h0_h1_h2_w0_w1_w2_global
-        // tensor
-        constexpr auto c_k_n_h0_h1_h2_w0_w1_w2_global_tensor_step_hacks = CGlobalStepHacks{};
-
-        constexpr auto c_k0_k1_n_h0_h1_h2_w0_w1_w2_thread_copy_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<KPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThread>{}));
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        ThreadwiseTensorSliceTransfer_v1r3<
-            FloatAcc,
-            FloatC,
-            decltype(c_k0_k1_n_h0_h1_h2_w0_w1_w2_thread_copy_desc),
-            decltype(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc),
-            Sequence<I1, KPerThread, I1, I1, I1, HoPerThread, I1, I1, WoPerThread>,
-            CThreadTransferSrcDstAccessOrder,
-            CThreadTransferSrcDstVectorDim,
-            CThreadTransferDstScalarPerVector,
-            CGlobalMemoryDataOperation,
-            1,
-            true>(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                  make_multi_index(k_block_work_id,
-                                   k_thread_data_on_global,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0))
-            .Run(c_k0_k1_n_h0_h1_h2_w0_w1_w2_thread_copy_desc,
-                 make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                 c_thread_buf,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                 c_global_buf,
-                 c_k_n_h0_h1_h2_w0_w1_w2_global_tensor_step_hacks);
-    }
-
-    template <typename CThreadBuff,
-              typename DGlobalBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename CThreadDesc_K1_N_H2_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx>
-    __device__ static void
-    MaxPool(const CThreadBuff& c_thread_buf,
-            DGlobalBuff& d_global_buf,
-            const CBlockIndex& c_block_idx,
-            const CThreadIndex& c_thread_idx,
-            const CThreadDesc_K1_N_H2_W2&,
-            const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc)
-    {
-
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        const auto k_thread_id  = c_thread_idx[I0];
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        static_assert(HoPerThread % 2 == 0 && WoPerThread % 2 == 0, "");
-
-        constexpr auto HoPerThread_2 = HoPerThread / 2;
-        constexpr auto WoPerThread_2 = WoPerThread / 2;
-
-        constexpr auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<KPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThread_2>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThread_2>{}));
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatC,
-                     d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.GetElementSpaceSize(),
-                     true>
-            d_thread_buf;
-
-        static_for<0, KPerThread, 1>{}([&](auto ki) {
-            static_for<0, HoPerThread_2, 1>{}([&](auto hi) {
-                static_for<0, WoPerThread_2, 1>{}([&](auto wi) {
-                    constexpr index_t d_offset =
-                        d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.CalculateOffset(
-                            make_tuple(0, ki, 0, 0, 0, hi, 0, 0, wi));
-
-                    constexpr index_t c_offset_0 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2, wi * 2));
-                    constexpr index_t c_offset_1 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2, wi * 2 + 1));
-                    constexpr index_t c_offset_2 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2 + 1, wi * 2));
-                    constexpr index_t c_offset_3 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2 + 1, wi * 2 + 1));
-
-                    d_thread_buf(Number<d_offset>{}) = c_thread_buf[Number<c_offset_0>{}];
-                    d_thread_buf(Number<d_offset>{}) =
-                        fmaxf(c_thread_buf[Number<c_offset_1>{}], d_thread_buf(Number<d_offset>{}));
-                    d_thread_buf(Number<d_offset>{}) =
-                        fmaxf(c_thread_buf[Number<c_offset_2>{}], d_thread_buf(Number<d_offset>{}));
-                    d_thread_buf(Number<d_offset>{}) =
-                        fmax(c_thread_buf[Number<c_offset_3>{}], d_thread_buf(Number<d_offset>{}));
-                });
-            });
-        });
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        constexpr auto d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks = DGlobalStepHacks{};
-
-        ThreadwiseTensorSliceTransfer_v1r3<
-            FloatC,
-            FloatC,
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc),
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc),
-            Sequence<I1, KPerThread, I1, I1, I1, HoPerThread_2, I1, I1, WoPerThread_2>,
-            CThreadTransferSrcDstAccessOrder,
-            CThreadTransferSrcDstVectorDim,
-            CThreadTransferDstScalarPerVector,
-            InMemoryDataOperationEnum::Set,
-            1,
-            true>(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                  make_multi_index(k_block_work_id,
-                                   k_thread_data_on_global,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0))
-            .Run(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc,
-                 make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                 d_thread_buf,
-                 d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                 d_global_buf,
-                 d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks);
-    }
-
-    template <typename CThreadBuff,
-              typename DGlobalBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename CThreadDesc_K1_N_H2_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx>
-    __device__ static void
-    ResizeAdd(const CThreadBuff& c_thread_buf,
-              DGlobalBuff& d_global_buf,
-              const CBlockIndex& c_block_idx,
-              const CThreadIndex& c_thread_idx,
-              const CThreadDesc_K1_N_H2_W2&,
-              const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc)
-    {
-
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        const auto k_thread_id  = c_thread_idx[I0];
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        constexpr auto HoPerThreadx2 = HoPerThread * 2;
-        constexpr auto WoPerThreadx2 = WoPerThread * 2;
-
-        constexpr auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<KPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThreadx2>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThreadx2>{}));
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatC,
-                     d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.GetElementSpaceSize(),
-                     true>
-            d_thread_buf;
-
-        static_for<0, KPerThread, 1>{}([&](auto k_i) {
-            static_for<0, HoPerThreadx2, 1>{}([&](auto h_i) {
-                static_for<0, WoPerThreadx2, 1>{}([&](auto w_i) {
-                    d_thread_buf(Number<d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.CalculateOffset(
-                                     make_tuple(0, k_i, 0, 0, 0, h_i, 0, 0, w_i))>{}) =
-                        c_thread_buf[Number<c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                            make_tuple(k_i, 0, h_i / 2, w_i / 2))>{}];
-                });
-            });
-        });
-
-        // hack to control index calculation when iterating over d_k_n_ho_wo_global tensor
-        constexpr auto d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks = DGlobalStepHacks{};
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        ThreadwiseTensorSliceTransfer_v1r3<
-            FloatC,
-            FloatC,
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc),
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc),
-            Sequence<I1, KPerThread, I1, I1, I1, HoPerThreadx2, I1, I1, WoPerThreadx2>,
-            CThreadTransferSrcDstAccessOrder,
-            CThreadTransferSrcDstVectorDim,
-            CThreadTransferDstScalarPerVector,
-            InMemoryDataOperationEnum::Add,
-            1,
-            true>(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                  make_multi_index(k_block_work_id,
-                                   k_thread_data_on_global,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0))
-            .Run(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc,
-                 make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                 d_thread_buf,
-                 d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                 d_global_buf,
-                 d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks);
-    }
-
-    template <typename AGlobalBuff,
-              typename BGlobalBuff,
-              typename CThreadBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CThreadDesc_K1_N_H2_W2,
-              bool HasMainE0BlockLoop>
-    __device__ static void
-    GemmOp(const AGlobalBuff& a_global_buf,
-           const BGlobalBuff& b_global_buf,
-           CThreadBuff& c_thread_buf,
-           FloatAB* __restrict__ p_shared_block,
-           const CBlockIndex& c_block_idx,
-           const CThreadIndex& c_thread_idx,
-           const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-           const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-           const CThreadDesc_K1_N_H2_W2&,
-           integral_constant<bool, HasMainE0BlockLoop>)
-    {
-        constexpr auto HasMainE1BlockLoop       = CalculateHasMainE1BlockLoop();
-        constexpr auto HasDoubleTailE1BlockLoop = CalculateHasDoubleTailE1BlockLoop();
-
-        // const auto c_k_n_h_w_block_cluster_idx =
-        // GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-        // cblockid_to_k_n_h_w_block_cluster_adaptor.CalculateBottomIndex(
-        // make_multi_index(get_block_1d_id()));
-
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        constexpr auto max_lds_align = Number<ABlockTransferDstScalarPerVector_E2>{};
-
-        constexpr auto a_e1_k1_e2_block_gemm_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E1PerBlock>{}, Number<KPerBlock>{}, Number<E2>{}), max_lds_align);
-
-        constexpr auto b_e1_n_h_w_e2_block_gemm_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(Number<E1PerBlock>{},
-                                                           I1,
-                                                           Number<HoPerBlock>{},
-                                                           Number<WoPerBlock>{},
-                                                           Number<E2>{}));
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v3<BlockSize,
-                                                 FloatAB,
-                                                 FloatAB,
-                                                 FloatAcc,
-                                                 decltype(a_e1_k1_e2_block_gemm_desc),
-                                                 decltype(b_e1_n_h_w_e2_block_gemm_desc),
-                                                 decltype(c_k1_n_h2_w2_thread_gemm_desc),
-                                                 EPerThread,
-                                                 K2>{};
-        // blockwise_gemm.GetBeginOfCThreadDesc_K_N_Ho_Wo(get_thread_local_1d_id());
-
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        constexpr auto a_e0_e1_k0_k1_e2_block_copy_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<I1>{}, Number<E1>{}, I1, Number<KPerBlock>{}, Number<E2>{}),
-            max_lds_align);
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<I1, E1, I1, KPerBlock, E2>,
-                                            ABlockTransferThreadSliceLengths_E0_E1_K0_K1_E2,
-                                            ABlockTransferThreadClusterLengths_E0_E1_K0_K1_E2,
-                                            ABlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(a_e0_e1_k0_k1_e2_grid_desc),
-                                            decltype(a_e0_e1_k0_k1_e2_block_copy_desc),
-                                            ABlockTransferSrcAccessOrder,
-                                            Sequence<0, 1, 2, 3, 4>,
-                                            ABlockTransferSrcVectorDim,
-                                            4,
-                                            ABlockTransferSrcScalarPerVector,
-                                            ABlockTransferDstScalarPerVector_E2,
-                                            1,
-                                            1,
-                                            AThreadTransferSrcResetCoordinateAfterRun,
-                                            false>(a_e0_e1_k0_k1_e2_grid_desc,
-                                                   make_multi_index(0, 0, k_block_work_id, 0, 0),
-                                                   a_e0_e1_k0_k1_e2_block_copy_desc,
-                                                   make_multi_index(0, 0, 0, 0, 0));
-
-        constexpr auto a_block_slice_copy_step = make_multi_index(I1, 0, 0, 0, 0);
-
-        constexpr auto b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<E1PerBlock>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThread>{},
-                                                           Number<E2>{}));
-
-        auto b_threadwise_transfer = ThreadwiseTensorSliceTransfer_v2<
-            FloatAB,
-            FloatAB,
-            decltype(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc),
-            decltype(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc),
-            Sequence<I1, E1PerBlock, I1, I1, I1, HoPerThread, I1, I1, WoPerThread, E2>,
-            BBlockTransferSrcAccessOrder,
-            BBlockTransferSrcVectorDim,
-            BBlockTransferSrcScalarPerVector,
-            BThreadTransferSrcResetCoordinateAfterRun,
-            true>(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                  make_multi_index(0,
-                                   0,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0,
-                                   0));
-
-        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_shared_block, a_e0_e1_k0_k1_e2_block_copy_desc.GetElementSpaceSize());
-
-        //// register allocation for output
-        // StaticBuffer<AddressSpaceEnum::Vgpr,
-        // FloatAcc,
-        // c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-        // true>
-        // c_thread_buf;
-
-        // initialize output thread tensor
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_k1_n_h2_w2_thread_gemm_desc),
-                                    Sequence<KPerThread, I1, HoPerThread, WoPerThread>>{}
-            .Run(c_k1_n_h2_w2_thread_gemm_desc,
-                 make_tuple(I0, I0, I0, I0),
-                 c_thread_buf,
-                 FloatAcc{0});
-
-        constexpr auto b_thread_slice_copy_step =
-            make_multi_index(0, E1PerBlock, 0, 0, 0, 0, 0, 0, 0, 0);
-
-        // hack to control index calculation when iterating over A and B matrix for threadwise copy
-        constexpr auto a_e0_e1_k_e2_global_step_hacks                   = AGlobalStepHacks{};
-        constexpr auto b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks = BGlobalStepHacks{};
-
-        // double regsiter buffer for b
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAB,
-                     b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc.GetElementSpaceSize(),
-                     true>
-            b_thread_even_buf, b_thread_odd_buf;
-
-        if constexpr(HasMainE0BlockLoop)
-        {
-            const auto E0 = b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetLength(I0);
-
-            index_t e0_block_data_begin = 0;
-
-            do
-            {
-                // LDS double buffer: preload data
-                {
-                    a_blockwise_copy.RunRead(
-                        a_e0_e1_k0_k1_e2_grid_desc, a_global_buf, a_e0_e1_k_e2_global_step_hacks);
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_even_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    a_blockwise_copy.RunWrite(a_e0_e1_k0_k1_e2_block_copy_desc, a_block_buf);
-                }
-
-                __syncthreads();
-
-                if constexpr(HasMainE1BlockLoop)
-                {
-                    index_t e1_block_data_begin = 0;
-
-                    // LDS double buffer: main body
-                    // use Do-While loop instead of For loop to simplify control flow
-                    do
-                    {
-                        // even iteration
-                        b_threadwise_transfer.MoveSrcSliceWindow(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_thread_slice_copy_step,
-                            BGlobalMoveSliceWindowStepHacks{});
-
-                        b_threadwise_transfer.Run(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_global_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                            make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                            b_thread_odd_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                        // LDS double buffer: GEMM on current data
-                        blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                        blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                        b_threadwise_transfer.MoveSrcSliceWindow(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_thread_slice_copy_step,
-                            BGlobalMoveSliceWindowStepHacks{});
-
-                        b_threadwise_transfer.Run(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_global_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                            make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                            b_thread_even_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                        // LDS double buffer: GEMM on current data
-                        blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-
-                        blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                        e1_block_data_begin += 2 * E1PerBlock;
-
-                    } while(e1_block_data_begin < E1 - 2 * E1PerBlock);
-                }
-
-                // LDS double buffer: tail
-                if constexpr(HasDoubleTailE1BlockLoop) // if has 2 iteration left
-                {
-                    b_threadwise_transfer.MoveSrcSliceWindow(
-                        b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                        b_thread_slice_copy_step,
-                        BGlobalMoveSliceWindowStepHacks{});
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_odd_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    // LDS double buffer: GEMM on 2nd-last data
-                    blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                    blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                    // LDS double buffer: GEMM on last data
-                    blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-                }
-                else // if has 1 iteration left
-                {
-                    // LDS double buffer: GEMM on last data
-                    blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-                }
-
-                a_blockwise_copy.MoveSrcSliceWindow(a_e0_e1_k0_k1_e2_grid_desc,
-                                                    a_block_slice_copy_step,
-                                                    AGlobalMoveSliceWindowStepHacks{});
-
-                blockwise_gemm.MoveABlockSliceWindow(make_tuple(-(E1 - E1PerBlock), 0, 0));
-
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                                         b_thread_slice_copy_step,
-                                                         BGlobalMoveSliceWindowStepHacks{});
-
-                e0_block_data_begin += 1;
-
-            } while(e0_block_data_begin < E0);
-        }
-        else
-        {
-            // LDS double buffer: preload data
-            {
-                a_blockwise_copy.RunRead(
-                    a_e0_e1_k0_k1_e2_grid_desc, a_global_buf, a_e0_e1_k_e2_global_step_hacks);
-
-                b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                          b_global_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                          make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                          b_thread_even_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                a_blockwise_copy.RunWrite(a_e0_e1_k0_k1_e2_block_copy_desc, a_block_buf);
-            }
-
-            __syncthreads();
-
-            if constexpr(HasMainE1BlockLoop)
-            {
-                index_t e1_block_data_begin = 0;
-
-                // LDS double buffer: main body
-                // use Do-While loop instead of For loop to simplify control flow
-                do
-                {
-                    // even iteration
-                    b_threadwise_transfer.MoveSrcSliceWindow(
-                        b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                        b_thread_slice_copy_step,
-                        BGlobalMoveSliceWindowStepHacks{});
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_odd_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    // LDS double buffer: GEMM on current data
-                    blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                    blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                    b_threadwise_transfer.MoveSrcSliceWindow(
-                        b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                        b_thread_slice_copy_step,
-                        BGlobalMoveSliceWindowStepHacks{});
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_even_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    // LDS double buffer: GEMM on current data
-                    blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-
-                    blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                    e1_block_data_begin += 2 * E1PerBlock;
-
-                } while(e1_block_data_begin < E1 - 2 * E1PerBlock);
-            }
-
-            // LDS double buffer: tail
-            if constexpr(HasDoubleTailE1BlockLoop) // if has 2 iteration left
-            {
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                                         b_thread_slice_copy_step,
-                                                         BGlobalMoveSliceWindowStepHacks{});
-
-                b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                          b_global_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                          make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                          b_thread_odd_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                // LDS double buffer: GEMM on 2nd-last data
-                blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                // LDS double buffer: GEMM on last data
-                blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-            }
-            else // if has 1 iteration left
-            {
-                // LDS double buffer: GEMM on last data
-                blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-            }
-        }
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop>
-    __device__ static void
-    Conv(const FloatAB* __restrict__ p_a_global,
-         const FloatAB* __restrict__ p_b_global,
-         const FloatC* __restrict__ p_bias_global,
-         FloatC* __restrict__ p_c_global,
-         FloatC* __restrict__ p_d_global,
-         FloatAB* __restrict__ p_shared_block,
-         const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-         const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-         const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-         const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-         const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-         integral_constant<bool, HasMainE0BlockLoop>)
-    {
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetElementSpaceSize());
-        auto d_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_d_global, d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Output
-        WriteOut(c_thread_buf,
-                 c_global_buf,
-                 c_k_n_h_w_block_cluster_idx,
-                 c_thread_mtx_index,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop,
-              ActivTypeEnum ActivType>
-    __device__ static void ConvBiasActiv(
-        const FloatAB* __restrict__ p_a_global,
-        const FloatAB* __restrict__ p_b_global,
-        const FloatC* __restrict__ p_bias_global,
-        FloatC* __restrict__ p_c_global,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-        const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-        integral_constant<bool, HasMainE0BlockLoop>,
-        integral_constant<ActivTypeEnum, ActivType>)
-    {
-        static constexpr auto activ_type = integral_constant<ActivTypeEnum, ActivType>{};
-
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Bias
-        BiasOp(bias_global_buf,
-               c_thread_buf,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               bias_k0_k1_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc);
-
-        // Activ
-        Activation(c_thread_buf, c_k1_n_h2_w2_thread_gemm_desc, activ_type);
-
-        // Output
-        WriteOut(c_thread_buf,
-                 c_global_buf,
-                 c_k_n_h_w_block_cluster_idx,
-                 c_thread_mtx_index,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop,
-              ActivTypeEnum ActivType>
-    __device__ static void ConvBiasActivMaxpool(
-        const FloatAB* __restrict__ p_a_global,
-        const FloatAB* __restrict__ p_b_global,
-        const FloatC* __restrict__ p_bias_global,
-        FloatC* __restrict__ p_c_global,
-        FloatC* __restrict__ p_d_global,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-        const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-        const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-        integral_constant<bool, HasMainE0BlockLoop>,
-        integral_constant<ActivTypeEnum, ActivType>)
-    {
-        static constexpr auto activ_type = integral_constant<ActivTypeEnum, ActivType>{};
-
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetElementSpaceSize());
-        auto d_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_d_global, d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Bias
-        BiasOp(bias_global_buf,
-               c_thread_buf,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               bias_k0_k1_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc);
-
-        // Activ
-        Activation(c_thread_buf, c_k1_n_h2_w2_thread_gemm_desc, activ_type);
-
-        // Output
-        WriteOut(c_thread_buf,
-                 c_global_buf,
-                 c_k_n_h_w_block_cluster_idx,
-                 c_thread_mtx_index,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        // MaxPool
-        MaxPool(c_thread_buf,
-                d_global_buf,
-                c_k_n_h_w_block_cluster_idx,
-                c_thread_mtx_index,
-                c_k1_n_h2_w2_thread_gemm_desc,
-                d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc);
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop,
-              ActivTypeEnum ActivType>
-    __device__ static void ConvBiasActivResizeAdd(
-        const FloatAB* __restrict__ p_a_global,
-        const FloatAB* __restrict__ p_b_global,
-        const FloatC* __restrict__ p_bias_global,
-        FloatC* __restrict__ p_d_global,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-        const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-        const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-        integral_constant<bool, HasMainE0BlockLoop>,
-        integral_constant<ActivTypeEnum, ActivType>)
-    {
-        static constexpr auto activ_type = integral_constant<ActivTypeEnum, ActivType>{};
-
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto d_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_d_global, d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Bias
-        BiasOp(bias_global_buf,
-               c_thread_buf,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               bias_k0_k1_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc);
-
-        // Activ
-        Activation(c_thread_buf, c_k1_n_h2_w2_thread_gemm_desc, activ_type);
-
-        // Resize_Add
-        ResizeAdd(c_thread_buf,
-                  d_global_buf,
-                  c_k_n_h_w_block_cluster_idx,
-                  c_thread_mtx_index,
-                  c_k1_n_h2_w2_thread_gemm_desc,
-                  d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc);
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index 578665ea85fb723565e597bbec26457ffb449606..7289a20da03523b0496353a0cea48e0568c50de0 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..119c1ea596dc6c7f3cff9b20a94c9f2dbe85e4d4
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
@@ -0,0 +1,956 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename GridwiseOp,
+          typename ADataType,
+          typename BDataType,
+          typename DsPointer,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AGridDesc_AK0_M_AK1,
+          typename BGridDesc_BK0_N_BK1,
+          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename Block2CTileMap,
+          typename ComputePtrOffsetOfBatch,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_grouped_conv_fwd_multiple_d_wmma_cshuffle(
+            const ADataType* __restrict__ p_a_grid,
+            const BDataType* __restrict__ p_b_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const index_t batch_count,
+            const AGridDesc_AK0_M_AK1 a_grid_desc_k0_m_k1,
+            const BGridDesc_BK0_N_BK1 b_grid_desc_k0_n_k1,
+            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+            const EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock_,
+            const Block2CTileMap block_2_ctile_map,
+            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__))
+    // offset base pointer for each work-group
+    const index_t num_blocks_per_batch =
+        __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+    const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
+
+    const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
+    const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
+    const long_index_t e_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetEPtrOffset(g_idx)));
+
+    const auto ds_batch_offset = compute_ptr_offset_of_batch.GetDsPtrOffset(g_idx);
+
+    __shared__ char p_shared[GridwiseOp::GetSharedMemoryNumberOfByte()];
+
+    DsPointer p_ds_grid_grp;
+
+    static constexpr index_t NumDTensor =
+        DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock::Size();
+
+    static_for<0, NumDTensor, 1>{}(
+        [&](auto i) { p_ds_grid_grp(i) = p_ds_grid[i] + ds_batch_offset[i]; });
+
+    GridwiseOp::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
+                                                p_b_grid + b_batch_offset,
+                                                p_ds_grid_grp,
+                                                p_e_grid + e_batch_offset,
+                                                p_shared,
+                                                a_grid_desc_k0_m_k1,
+                                                b_grid_desc_k0_n_k1,
+                                                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                e_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                                a_element_op,
+                                                b_element_op,
+                                                cde_element_op,
+                                                block_2_ctile_map);
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = batch_count;
+    ignore = a_grid_desc_k0_m_k1;
+    ignore = b_grid_desc_k0_n_k1;
+    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock_;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = compute_ptr_offset_of_batch;
+    ignore = block_2_ctile_map;
+#endif
+}
+
+template <typename GridwiseOp,
+          typename ADataType,
+          typename BDataType,
+          typename DsPointer,
+          typename EDataType,
+          typename AGridDesc_K0_M_K1,
+          typename BGridDesc_K0_N_K1,
+          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename ComputePtrOffsetOfBatch,
+          typename Block2CTileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_contraction_multiple_d_wmma_cshuffle(
+            const ADataType* __restrict__ p_a_grid,
+            const BDataType* __restrict__ p_b_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const index_t batch_count,
+            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
+            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
+            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
+            const Block2CTileMap block_2_etile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__))
+    // printf("entry kernel launch");
+    __shared__ char p_shared[GridwiseOp::GetSharedMemoryNumberOfByte()];
+
+    const index_t num_blocks_per_batch =
+        __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+    const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
+
+    // printf("before compute_ptr_offset call");
+    const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
+    const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
+    const long_index_t e_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetEPtrOffset(g_idx)));
+
+    const auto ds_batch_offset = compute_ptr_offset_of_batch.GetDsPtrOffset(g_idx);
+
+    static constexpr index_t NumDTensor =
+        DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock::Size();
+
+    DsPointer p_ds_grid_grp;
+
+    // printf("before allocate pointer d");
+
+    static_for<0, NumDTensor, 1>{}(
+        [&](auto i) { p_ds_grid_grp(i) = p_ds_grid[i] + ds_batch_offset[i]; });
+
+    // printf("before entry");
+
+    GridwiseOp::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
+                                                p_b_grid + b_batch_offset,
+                                                p_ds_grid_grp,
+                                                p_e_grid + e_batch_offset,
+                                                p_shared,
+                                                a_grid_desc_k0_m_k1,
+                                                b_grid_desc_k0_n_k1,
+                                                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                a_element_op,
+                                                b_element_op,
+                                                cde_element_op,
+                                                block_2_etile_map);
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = batch_count;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = a_grid_desc_k0_m_k1;
+    ignore = b_grid_desc_k0_n_k1;
+    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = block_2_etile_map;
+    ignore = compute_ptr_offset_of_batch;
+#endif
+}
+
+template <typename GridwiseOp,
+          typename ADataType,
+          typename BDataType,
+          typename DsPointer,
+          typename EDataType,
+          typename AGridDesc_K0_M_K1,
+          typename BGridDesc_K0_N_K1,
+          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename Block2CTileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_mupltipe_d_wmma_cshuffle(
+            const ADataType* __restrict__ p_a_grid,
+            const BDataType* __restrict__ p_b_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
+            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
+            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const Block2CTileMap block_2_ctile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__))
+    __shared__ char p_shared[GridwiseOp::GetSharedMemoryNumberOfByte()];
+
+    GridwiseOp::template Run<HasMainKBlockLoop>(p_a_grid,
+                                                p_b_grid,
+                                                p_ds_grid,
+                                                p_e_grid,
+                                                p_shared,
+                                                a_grid_desc_k0_m_k1,
+                                                b_grid_desc_k0_n_k1,
+                                                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                a_element_op,
+                                                b_element_op,
+                                                cde_element_op,
+                                                block_2_ctile_map);
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = a_grid_desc_k0_m_k1;
+    ignore = b_grid_desc_k0_n_k1;
+    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = block_2_ctile_map;
+#endif // end of if (defined(__gfx1100__ ))
+}
+
+template < // DataType Family
+    typename ADataType,
+    typename BDataType,
+    typename AccDataType,
+    typename CShuffleDataType,
+    typename DsDataType,
+    typename EDataType,
+    // InMemory Data Descriptor
+    typename AGridDesc_K0_M_K1,
+    typename BGridDesc_K0_N_K1,
+    typename DsGridDesc_M_N,
+    typename EGridDesc_M_N,
+    // ElementwiseOp Family
+    typename AElementwiseOperation,
+    typename BElementwiseOperation,
+    typename CDEElementwiseOperation,
+    InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+    // Tiling Family
+    index_t MPerBlock,
+    index_t NPerBlock,
+    index_t K0PerBlock,
+    index_t MPerWmma,
+    index_t NPerWmma,
+    index_t K1Value,
+    index_t MRepeat,
+    index_t NRepeat,
+    // ThreadCluster Family
+    index_t BlockSize,
+    typename ABlockTransferThreadClusterLengths_K0_M_K1,
+    typename ABlockTransferThreadClusterArrangeOrder,
+    typename ABlockTransferSrcAccessOrder,
+    index_t ABlockTransferSrcVectorDim,
+    index_t ABlockTransferSrcScalarPerVector,
+    index_t ABlockTransferDstScalarPerVector_K1,
+    bool AThreadTransferSrcResetCoordinateAfterRun,
+    bool ABlockLdsExtraM,
+    typename BBlockTransferThreadClusterLengths_K0_N_K1,
+    typename BBlockTransferThreadClusterArrangeOrder,
+    typename BBlockTransferSrcAccessOrder,
+    index_t BBlockTransferSrcVectorDim,
+    index_t BBlockTransferSrcScalarPerVector,
+    index_t BBlockTransferDstScalarPerVector_K1,
+    bool BThreadTransferSrcResetCoordinateAfterRun,
+    bool BBlockLdsExtraN,
+    index_t CShuffleMRepeatPerShuffle,
+    index_t CShuffleNRepeatPerShuffle,
+    typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+    index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+    index_t NumGemmKPrefetchStage = 1,
+    LoopScheduler LoopSched       = make_default_loop_scheduler(),
+    PipelineVersion PipelineVer   = PipelineVersion::v1>
+struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto K1 = Number<K1Value>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    using GridwiseGemmPipe = remove_cvref_t<decltype(
+        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
+    {
+        constexpr auto max_lds_align = K1;
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_k0perblock_mperblock_k1 = [&]() {
+            if constexpr(ABlockLdsExtraM)
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1),
+                    make_tuple(Number<MPerBlock + 1>{} * K1, K1, I1));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_aligned(
+                    make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1), max_lds_align);
+            }
+        }();
+
+        return a_block_desc_k0perblock_mperblock_k1;
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_K0PerBlock_NPerBlock_K1()
+    {
+        constexpr auto max_lds_align = K1;
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_k0perblock_nperblock_k1 = [&]() {
+            if constexpr(BBlockLdsExtraN)
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1),
+                    make_tuple(Number<NPerBlock + 1>{} * K1, K1, I1));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_aligned(
+                    make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1), max_lds_align);
+            }
+        }();
+
+        return b_block_desc_k0perblock_nperblock_k1;
+    }
+
+    __host__ __device__ static constexpr auto
+    // *Caution Here repeat is shuffle repeat
+    GetCShuffleBlockDescriptor_MShRepeat_MPerShRepeat_NShRepeat_NPerShRepeat()
+    {
+        constexpr index_t MWave = MPerBlock / (MRepeat * MPerWmma);
+        constexpr index_t NWave = NPerBlock / (NRepeat * NPerWmma);
+
+        constexpr auto c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMRepeatPerShuffle * MWave * MPerWmma>{},
+                           I1,
+                           Number<CShuffleNRepeatPerShuffle * NWave * NPerWmma>{}));
+
+        return c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat;
+    }
+
+    // ck::Tuple<const D0DataType*, const D1DataType*, ...>
+    static constexpr auto MakeDsGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                return static_cast<const DDataType*>(nullptr);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_k0perblock_mperblock_k1 =
+            GetABlockDescriptor_K0PerBlock_MPerBlock_K1();
+
+        constexpr auto b_block_desc_k0perblock_nperblock_k1 =
+            GetBBlockDescriptor_K0PerBlock_NPerBlock_K1();
+
+        constexpr auto cshuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat =
+            GetCShuffleBlockDescriptor_MShRepeat_MPerShRepeat_NShRepeat_NPerShRepeat();
+
+        constexpr auto max_lds_align = K1;
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_k0perblock_mperblock_k1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_k0perblock_nperblock_k1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto c_block_space_size_aligned = math::integer_least_multiple(
+            cshuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat.GetElementSpaceSize(),
+            max_lds_align);
+
+        return math::max((a_block_space_size_aligned * sizeof(ADataType) +
+                          b_block_space_size_aligned * sizeof(BDataType)),
+                         c_block_space_size_aligned * sizeof(CShuffleDataType));
+    }
+
+    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
+    template <typename Block2CTileMap>
+    __host__ __device__ static constexpr bool
+    CheckValidity(const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
+                  const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
+                  const DsGridDesc_M_N& ds_grid_desc_m_n,
+                  const EGridDesc_M_N& e_grid_desc_m_n,
+                  const Block2CTileMap& block_2_ctile_map)
+    {
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
+                      "wrong! K1 need to be known at compile-time");
+
+        static_assert((MPerBlock % (MPerWmma * MRepeat) == 0) &&
+                          (NPerBlock % (NRepeat * NPerWmma)) == 0,
+                      "Invalid tuning param!");
+
+        const auto M  = a_grid_desc_k0_m_k1.GetLength(I1);
+        const auto N  = b_grid_desc_k0_n_k1.GetLength(I1);
+        const auto K0 = a_grid_desc_k0_m_k1.GetLength(I0);
+
+        bool valid = true;
+
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            valid = valid && (M == ds_grid_desc_m_n[i].GetLength(I0) &&
+                              N == ds_grid_desc_m_n[i].GetLength(I1));
+        });
+
+        if(!valid)
+        {
+            return false;
+        }
+
+        if(!(M == e_grid_desc_m_n.GetLength(I0) && N == e_grid_desc_m_n.GetLength(I1) &&
+             K0 == b_grid_desc_k0_n_k1.GetLength(I0) && K1 == a_grid_desc_k0_m_k1.GetLength(I2) &&
+             K1 == b_grid_desc_k0_n_k1.GetLength(I2)))
+            return false;
+
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K0 % K0PerBlock == 0))
+            return false;
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = K0 / K0PerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        if(!block_2_ctile_map.CheckValidity(e_grid_desc_m_n))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        constexpr long_index_t TwoGB = (long_index_t{1} << 31);
+
+        if(!(a_grid_desc_k0_m_k1.GetElementSpaceSize() * sizeof(ADataType) <= TwoGB &&
+             b_grid_desc_k0_n_k1.GetElementSpaceSize() * sizeof(BDataType) <= TwoGB &&
+             e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EDataType) <= TwoGB))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / (K0PerBlock * K1);
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    // E desc for destination in blockwise copy
+    template <typename EGridDesc_M_N_>
+    __host__ __device__ static constexpr auto
+    MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const EGridDesc_M_N_& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            e_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return e_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // Ds desc for source in blockwise copy
+    template <typename DsGridDesc_M_N_>
+    __host__ __device__ static constexpr auto
+    MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const DsGridDesc_M_N_& ds_grid_desc_m_n)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // return block_id to C matrix tile idx (m0, n0) mapping
+    __host__ __device__ static constexpr auto MakeDefaultBlock2CTileMap(
+        const EGridDesc_M_N& e_grid_desc_m_n, index_t /* M01 */, index_t /* N01 */)
+    {
+        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, EGridDesc_M_N>(
+            e_grid_desc_m_n);
+    }
+
+    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
+        MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
+        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using DefaultBlock2CTileMap =
+        remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(EGridDesc_M_N{}, 1, 1))>;
+    using DsGridPointer = decltype(MakeDsGridPointer());
+
+    template <bool HasMainKBlockLoop, typename Block2CTileMap = DefaultBlock2CTileMap>
+    __device__ static void Run(const ADataType* __restrict__ p_a_grid,
+                               const BDataType* __restrict__ p_b_grid,
+                               DsGridPointer p_ds_grid,
+                               EDataType* __restrict__ p_e_grid,
+                               void* __restrict__ p_shared,
+                               const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
+                               const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
+                               const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   e_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation& cde_element_op,
+                               const Block2CTileMap& block_2_ctile_map)
+    {
+        // printf("safe entry");
+        // clang-format off
+/*******************************************************************************/
+// Memory buffer zone.
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_k0_m_k1.GetElementSpaceSize());
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_k0_n_k1.GetElementSpaceSize());
+        const auto ds_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_ds_grid[i],
+                    ds_grid_desc_mblock_mperblock_nblock_nperblock[i].GetElementSpaceSize());
+            },
+            Number<NumDTensor>{});
+        auto e_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_e_grid, e_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+/*******************************************************************************/
+// BlockIdx.x -> [BlockId.m, BlockId.n]
+        const auto block_work_idx = block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+        if(!block_2_ctile_map.ValidCTileIndex(
+               block_work_idx,
+               make_tuple(e_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
+                          e_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
+        { return; }
+
+        // Store BlockId into SGPR
+        const index_t m_block_data_idx_on_grid = __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+        const index_t n_block_data_idx_on_grid = __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
+
+/*******************************************************************************/
+// BlockLevel, A/B Matrix ThreadMapping in LDS, As Destinaion of BlockWise_Copy
+        const auto K0 = a_grid_desc_k0_m_k1.GetLength(I0);
+        constexpr auto max_lds_align = K1;
+        constexpr auto a_block_desc_k0perblock_mperblock_k1 = GetABlockDescriptor_K0PerBlock_MPerBlock_K1();
+        constexpr auto b_block_desc_k0perblock_nperblock_k1 = GetBBlockDescriptor_K0PerBlock_NPerBlock_K1();
+        // A matrix blockwise copy
+        auto a_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<        ThisThreadBlock,
+/* typename SrcElementwiseOperation,              */    AElementwiseOperation,
+/* typename DstElementwiseOperation,              */    ck::tensor_operation::element_wise::PassThrough,
+/* InMemoryDataOperationEnum DstInMemOp,          */    InMemoryDataOperationEnum::Set,
+/* typename BlockSliceLengths,                    */    Sequence<K0PerBlock, MPerBlock, K1>,
+/* typename ThreadClusterLengths,                 */    ABlockTransferThreadClusterLengths_K0_M_K1,
+/* typename ThreadClusterArrangeOrder,            */    ABlockTransferThreadClusterArrangeOrder,
+/* typename SrcData,                              */    ADataType,
+/* typename DstData,                              */    ADataType,
+/* typename SrcDesc,                              */    decltype(a_grid_desc_k0_m_k1),
+/* typename DstDesc,                              */    decltype(a_block_desc_k0perblock_mperblock_k1),
+/* typename SrcDimAccessOrder,                    */    ABlockTransferSrcAccessOrder,
+/* typename DstDimAccessOrder,                    */    Sequence<0, 1, 2>,
+/* index_t SrcVectorDim,                          */    ABlockTransferSrcVectorDim,
+/* index_t DstVectorDim,                          */    2,
+/* index_t SrcScalarPerVector,                    */    ABlockTransferSrcScalarPerVector,
+/* index_t DstScalarPerVector,                    */    ABlockTransferDstScalarPerVector_K1,
+/* index_t SrcScalarStrideInVector,               */    1,
+/* index_t DstScalarStrideInVector,               */    1,
+/* bool ThreadTransferSrcResetCoordinateAfterRun, */    AThreadTransferSrcResetCoordinateAfterRun,
+/* bool ThreadTransferDstResetCoordinateAfterRun, */    true>(
+                a_grid_desc_k0_m_k1,
+                make_multi_index(0, m_block_data_idx_on_grid, 0),
+                a_element_op,
+                a_block_desc_k0perblock_mperblock_k1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // B matrix blockwise copy
+        auto b_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<K0PerBlock, NPerBlock, K1>,
+                                                BBlockTransferThreadClusterLengths_K0_N_K1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                BDataType,
+                                                BDataType,
+                                                decltype(b_grid_desc_k0_n_k1),
+                                                decltype(b_block_desc_k0perblock_nperblock_k1),
+                                                BBlockTransferSrcAccessOrder,
+                                                Sequence<0, 1, 2>,
+                                                BBlockTransferSrcVectorDim,
+                                                2,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_K1,
+                                                1,
+                                                1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                true>(
+                b_grid_desc_k0_n_k1,
+                make_multi_index(0, n_block_data_idx_on_grid, 0),
+                b_element_op,
+                b_block_desc_k0perblock_nperblock_k1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+/*******************************************************************************/
+        // GEMM
+        constexpr auto WmmaK = 16;
+        constexpr auto KPack = math::integer_least_multiple(K1, WmmaK);
+
+        auto blockwise_gemm =
+            BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle<BlockSize,
+                                                         ADataType,
+                                                         BDataType,
+                                                         AccDataType,
+                                                         decltype(a_block_desc_k0perblock_mperblock_k1),
+                                                         decltype(b_block_desc_k0perblock_nperblock_k1),
+                                                         MPerWmma,
+                                                         NPerWmma,
+                                                         MRepeat,
+                                                         NRepeat,
+                                                         KPack>{};
+
+        // Prepare Register for C matrix
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+/*******************************************************************************/
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(a_block_desc_k0perblock_mperblock_k1.GetElementSpaceSize(), max_lds_align);
+        // LDS allocation for A and B: be careful of alignment
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(static_cast<ADataType*>(p_shared), a_block_desc_k0perblock_mperblock_k1.GetElementSpaceSize());
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(static_cast<BDataType*>(p_shared) + a_block_space_size_aligned, b_block_desc_k0perblock_nperblock_k1.GetElementSpaceSize());
+        
+        // Shift Per SUB_K
+        constexpr auto a_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
+
+        // gridwise GEMM pipeline
+        const index_t K0BlockMainLoop = __builtin_amdgcn_readfirstlane(K0 / K0PerBlock);
+        GridwiseGemmPipe::template Run<HasMainKBlockLoop>(a_grid_desc_k0_m_k1,
+                                                          a_block_desc_k0perblock_mperblock_k1,
+                                                          a_blockwise_copy,
+                                                          a_grid_buf,
+                                                          a_block_buf,
+                                                          a_block_slice_copy_step,
+                                                          b_grid_desc_k0_n_k1,
+                                                          b_block_desc_k0perblock_nperblock_k1,
+                                                          b_blockwise_copy,
+                                                          b_grid_buf,
+                                                          b_block_buf,
+                                                          b_block_slice_copy_step,
+                                                          blockwise_gemm,
+                                                          c_thread_buf,
+                                                          K0BlockMainLoop);
+/*******************************************************************************/
+        // write out to C, implement shuffle
+        {
+            constexpr auto c_thread_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs =  
+            blockwise_gemm.GetCThreadDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs();
+
+            // This API Provide All dimension (size) you need
+            constexpr auto c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp =
+                blockwise_gemm.GetCBlockDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs();
+
+            constexpr auto MWave              = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I1);
+            constexpr auto MSubGroup          = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I2);
+            constexpr auto NWave              = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I4);
+            constexpr auto NThreadPerSubGroup = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I5);
+            constexpr auto MAccVgprs          = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I6);
+
+            // LDS descriptor, shuffle and write out in MRepeat x NRepeat times
+            constexpr auto c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat =
+                GetCShuffleBlockDescriptor_MShRepeat_MPerShRepeat_NShRepeat_NPerShRepeat();
+
+            auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                static_cast<CShuffleDataType*>(p_shared),
+                c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat.GetElementSpaceSize());
+
+            constexpr auto c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs = transform_tensor_descriptor(
+                c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleMRepeatPerShuffle>{}, // MRepeat per shuffle repeat
+                        MWave,                               // MWave
+                        MSubGroup,                           // MSubGroup * MAccVgprs = MPerWmma
+                        MAccVgprs)),
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleNRepeatPerShuffle>{}, // NRepeat per shuffle repeat
+                        NWave,                               // NWave
+                        NThreadPerSubGroup))),               // NThreadPerSubGroup = NPerWmma
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<>{}, Sequence<0, 1, 2, 6>{}, Sequence<>{}, Sequence<3, 4, 5>{}));
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block = blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0);
+
+            const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+            const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_block_to_mrepeat_mwave_msubgroup_maccvgprs_adaptor =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_merge_transform(make_tuple(MRepeat, MWave, MSubGroup, MAccVgprs))),
+                make_tuple(Sequence<0, 1, 2, 3>{}),
+                make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_block_to_nrepeat_nwave_nthreadpersubgroup_adaptor =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_merge_transform(make_tuple(NRepeat, NWave, NThreadPerSubGroup))),
+                make_tuple(Sequence<0, 1, 2>{}),
+                make_tuple(Sequence<0>{}));
+            
+            const auto m_thread_data_on_block_idx = m_thread_data_on_block_to_mrepeat_mwave_msubgroup_maccvgprs_adaptor.CalculateBottomIndex(
+                make_multi_index(m_thread_data_on_block));
+            
+            const auto n_thread_data_on_block_idx = n_thread_data_on_block_to_nrepeat_nwave_nthreadpersubgroup_adaptor.CalculateBottomIndex(
+                make_multi_index(n_thread_data_on_block));
+
+            // shuffle: threadwise copy C from VGPR to LDS
+            auto c_thread_copy_vgpr_to_lds =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   CShuffleDataType,
+                                                   decltype(c_thread_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs),
+                                                   decltype(c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs),
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Sequence<CShuffleMRepeatPerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            CShuffleNRepeatPerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            MAccVgprs>,
+                                                   Sequence<0, 1, 2, 3, 4, 5, 6>,
+                                                   6,
+                                                   1, // vector write pixel
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>{
+                    c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs,
+                    make_multi_index(0,
+                                     m_thread_data_on_block_idx[I1],
+                                     m_thread_data_on_block_idx[I2],
+                                     0,
+                                     n_thread_data_on_block_idx[I1],
+                                     n_thread_data_on_block_idx[I2],
+                                     m_thread_data_on_block_idx[I3]),
+                    ck::tensor_operation::element_wise::PassThrough{}};
+            
+            // tuple of reference to C/Ds tensor descriptors
+            const auto c_ds_desc_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_desc_mblock_mperblock_nblock_nperblock[i]; },
+                    Number<NumDTensor>{}));
+            
+            // tuple of reference to C/Ds tensor buffers
+            const auto c_ds_buf_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_buf),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_buf[i]; },
+                    Number<NumDTensor>{}));
+
+            // tuple of starting index of C/Ds blockwise copy
+            const auto idx_c_ds_block_begin = container_concat(
+                make_tuple(make_multi_index(0, 0, 0, 0)),
+                generate_tuple(
+                    [&](auto) {
+                        return make_multi_index(block_work_idx[I0], 0, block_work_idx[I1], 0);
+                    },
+                    Number<NumDTensor>{}));
+
+            // shuffle: blockwise copy C from LDS to global
+            auto cde_shuffle_block_copy_lds_to_global = ThreadGroupTensorSliceTransfer_v7<
+                ThisThreadBlock,            // ThreadGroup
+                decltype(container_concat(make_tuple(CShuffleDataType{}), DsDataType{})),
+                Tuple<EDataType>,
+                decltype(c_ds_desc_refs),
+                decltype(tie(e_grid_desc_mblock_mperblock_nblock_nperblock)),
+                CDEElementwiseOperation,    // ElementwiseOperation,
+                Sequence<static_cast<index_t>(EGlobalMemoryDataOperation)>, // DstInMemOp,
+                Sequence<1,
+                         CShuffleMRepeatPerShuffle * MWave * MPerWmma,
+                         1,
+                         CShuffleNRepeatPerShuffle * NWave * NPerWmma>, // BlockSliceLengths,
+                CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                Sequence<0, 1, 2, 3>,                           // typename DimAccessOrder,
+                3,                                              // index_t VectorDim,
+                CDEShuffleBlockTransferScalarPerVector_NPerBlock, // index_t ScalarPerVector,
+                sequence_merge_t<
+                    Sequence<true>,
+                    uniform_sequence_gen_t<NumDTensor,
+                                           false>>,   // bool ThreadTransferSrcResetCoordinateAfterRun,
+                Sequence<false>> // bool ThreadTransferDstResetCoordinateAfterRun>
+                {c_ds_desc_refs,
+                 idx_c_ds_block_begin,
+                 tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                 make_tuple(make_multi_index(block_work_idx[I0], 0, block_work_idx[I1], 0)),
+                 cde_element_op};
+
+            // space filling curve for local reg & global memory
+            // space filling curve for threadwise C in VGPR
+            constexpr auto sfc_c_vgpr =
+                SpaceFillingCurve<Sequence<MRepeat, 1, 1, NRepeat, 1, 1, MAccVgprs>,
+                                  Sequence<0, 1, 2, 3, 4, 5, 6>,
+                                  Sequence<CShuffleMRepeatPerShuffle,
+                                           1,
+                                           1,
+                                           CShuffleNRepeatPerShuffle,
+                                           1,
+                                           1,
+                                           MAccVgprs>>{};
+
+            // space filling curve for shuffled blockwise C in global mem
+            constexpr auto sfc_cde_global =
+                SpaceFillingCurve<Sequence<1, MPerBlock, 1, NPerBlock>,
+                                  Sequence<0, 2, 1, 3>,
+                                  Sequence<1,
+                                           CShuffleMRepeatPerShuffle * MWave * MPerWmma,
+                                           1,
+                                           CShuffleNRepeatPerShuffle * NWave * NPerWmma>>{};
+
+            constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+            static_assert(num_access == sfc_cde_global.GetNumOfAccess(), "wrong!");
+
+            static_for<0, num_access, 1>{}([&](auto access_id) {
+                // make sure it's safe to write to LDS
+                block_sync_lds();
+
+                // each thread write its data from VGPR to LDS
+                c_thread_copy_vgpr_to_lds.Run(c_thread_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs,
+                                              sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                              c_thread_buf,
+                                              c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs,
+                                              c_shuffle_block_buf);
+
+                // make sure it's safe to read from LDS
+                block_sync_lds();
+
+                // each block copy its data from LDS to global
+                cde_shuffle_block_copy_lds_to_global.Run(
+                    c_ds_desc_refs,
+                    c_ds_buf_refs,
+                    tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                    tie(e_grid_buf));
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    constexpr auto cde_global_step = sfc_cde_global.GetForwardStep(access_id);
+                    // move on Ds
+                    static_for<0, NumDTensor, 1>{}([&](auto i) {
+                        cde_shuffle_block_copy_lds_to_global.MoveSrcSliceWindow(
+                            c_ds_desc_refs, i + I1, cde_global_step);
+                    });
+
+                    // move on E
+                    cde_shuffle_block_copy_lds_to_global.MoveDstSliceWindow(
+                        tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                        I0,
+                        cde_global_step);
+                }
+            });
+        }
+        // clang-format on
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index da0b0cea241d7b62dd8e138aa5dd7cb0ccdaa0cd..ec1cc53991d128a8f281e01289a6a8625810971a 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -92,6 +92,17 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
     using GridwiseGemmPipe = remove_cvref_t<decltype(
         GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
+    // denorm test fix, required to work around fp16 mfma issue
+    // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
+    // when mfma if fixed, remove this section and update
+    // ABDataTypeAdjusted -> ABDataType throughout this file
+#if CK_WORKAROUND_DENORM_FIX
+    using ABDataTypeAdjusted =
+        conditional_t<is_same_v<ABDataType, ck::half_t>, ck::bhalf_t, ABDataType>;
+#else
+    using ABDataTypeAdjusted = ABDataType;
+#endif
+
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
         // A matrix in LDS memory, dst of blockwise copy
@@ -397,7 +408,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
                                                 ABlockTransferThreadClusterLengths_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
                                                 ABDataType,
-                                                ABDataType,
+                                                ABDataTypeAdjusted,
                                                 decltype(a_grid_desc_ak0_m_ak1),
                                                 decltype(a_block_desc_ak0_m_ak1),
                                                 ABlockTransferSrcAccessOrder,
@@ -428,7 +439,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
                                                 BBlockTransferThreadClusterLengths_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
                                                 ABDataType,
-                                                ABDataType,
+                                                ABDataTypeAdjusted,
                                                 decltype(b_grid_desc_bk0_n_bk1),
                                                 decltype(b_block_desc_bk0_n_bk1),
                                                 BBlockTransferSrcAccessOrder,
@@ -458,11 +469,11 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         // sanity check
         constexpr index_t KPack =
             math::max(math::lcm(AK1, BK1),
-                      MfmaSelector<ABDataType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+                      MfmaSelector<ABDataTypeAdjusted, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            ABDataType,
+            ABDataTypeAdjusted,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
@@ -480,10 +491,11 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ABDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+            static_cast<ABDataTypeAdjusted*>(p_shared),
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ABDataType*>(p_shared) + a_block_space_size_aligned,
+            static_cast<ABDataTypeAdjusted*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
         constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
index 98331d854490abe4e7f478d817104218cb49c28f..f760feb2ed22015567ef66648d0d1afa81bdff9d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
@@ -1,8 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
+#include <iostream>
+
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp"
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp
index e9097552c2baf236b68370d6d98a771a2eec4e05..d1209636de93a13cbe7f851575b4abaedd99af72 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
index 3281b910d3d8782fc18f91ee77ce42aa44fc5ba9..d3d7d5af8577d1c2ed65fa817f708379c6089a25 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -79,6 +79,10 @@ struct GridwiseGemmPipeline_v2
 
             do
             {
+#if CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT
+                __builtin_amdgcn_iglp_opt(CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT);
+#endif
+
                 block_sync_lds();
 
                 // GEMM i
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
index 2fe55068449377a82b0c4bd12c65f05ea4e96199..3d1fbb73bd0087c682f9a72a2b6fb3d1f411cf62 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -54,7 +54,8 @@ __global__ void
             const ReduceGridDescriptor_MBlock_MPerBlock reduce_grid_desc_mblock_mperblock,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
index aa89bff9ee21873fdc57c839eb800be455005a9a..e7577bdcbd52b87cf986e0c9f92cc494089b9423 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..de5a4241986ff64e1797f2304da117178059ea4b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp
@@ -0,0 +1,157 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+
+namespace ck {
+
+template <typename TileLoadThreadGroup, index_t NumGemmKPrefetchStage>
+struct GridwiseGemmLoadWave;
+
+// 1-stage prefetch
+template <typename TileLoadThreadGroup>
+struct GridwiseGemmLoadWave<TileLoadThreadGroup, 1>
+{
+    __host__ __device__ static constexpr bool IsSupported(index_t /* num_loop */)
+    {
+        // TODO: improve applicability
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainLoop(index_t num_loop)
+    {
+        return num_loop > 1;
+    }
+
+    template <bool HasMainLoop,
+              typename AGridDesc,
+              typename ABlockDesc,
+              typename ABlockTransfer,
+              typename AGridBuffer,
+              typename ABlockBuffer,
+              typename ABlockTransferStep,
+              typename BGridDesc,
+              typename BBlockDesc,
+              typename BBlockTransfer,
+              typename BGridBuffer,
+              typename BBlockBuffer,
+              typename BBlockTransferStep>
+    static __device__ void RunLoadWavePipeline(const AGridDesc& a_grid_desc,
+                                               const ABlockDesc& a_block_desc,
+                                               ABlockTransfer& a_blockwise_copy,
+                                               const AGridBuffer& a_grid_buf,
+                                               ABlockBuffer& a_block_buf,
+                                               const ABlockTransferStep& a_block_copy_step,
+                                               const BGridDesc& b_grid_desc,
+                                               const BBlockDesc& b_block_desc,
+                                               BBlockTransfer& b_blockwise_copy,
+                                               const BGridBuffer& b_grid_buf,
+                                               BBlockBuffer& b_block_buf,
+                                               const BBlockTransferStep& b_block_copy_step,
+                                               index_t num_loop)
+    {
+        // global read 0
+        a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+        b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+        // move to 1
+        a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+        b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+        // LDS write 0
+        a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+        b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+        if constexpr(HasMainLoop)
+        {
+            index_t i = 0;
+
+            do
+            {
+                // sync for Load threads()
+                block_sync_lds();
+                // global read i + 1
+                a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+                b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+                // move to i + 2
+                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+                // sync with math threads()
+                block_sync_lds();
+
+                // LDS write i+1
+                a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+                b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+                ++i;
+            } while(i < (num_loop - 1));
+        }
+
+        // tail
+        {
+            block_sync_lds();
+            // GEMM num_loop - 1
+        }
+    }
+};
+
+template <typename TileMathThreadGroup, index_t NumGemmKPrefetchStage>
+struct GridwiseGemmMathWave;
+// 1- stage prefetch
+template <typename TileMathThreadGroup>
+struct GridwiseGemmMathWave<TileMathThreadGroup, 1>
+{
+
+    __host__ __device__ static constexpr bool IsSupported(index_t /* num_loop */) { return true; }
+
+    __host__ __device__ static constexpr bool CalculateHasMainLoop(index_t num_loop)
+    {
+        return num_loop > 1;
+    }
+
+    template <bool HasMainLoop,
+              typename ABlockBuffer,
+              typename BBlockBuffer,
+              typename BlockwiseGemm,
+              typename CThreadBuffer>
+    static __device__ void RunMathWavePipeline(ABlockBuffer& a_block_buf,
+                                               BBlockBuffer& b_block_buf,
+                                               const BlockwiseGemm& block_gemm,
+                                               CThreadBuffer& c_thread_buf,
+                                               index_t num_loop)
+    {
+        // Initialize C
+        c_thread_buf.Clear();
+
+        // main body
+        if constexpr(HasMainLoop)
+        {
+            index_t i = 0;
+
+            do
+            {
+                block_sync_lds();
+
+                // GEMM i
+                block_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
+
+                block_sync_lds();
+                ++i;
+            } while(i < (num_loop - 1));
+        }
+
+        // tail
+        {
+            block_sync_lds();
+
+            // GEMM num_loop - 1
+            block_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
+        }
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..740dde5c65f3d02282d6a0e5b05d41c6591de72f
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
@@ -0,0 +1,649 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemm,
+          typename FloatA,
+          typename FloatB,
+          typename FloatC,
+          typename AGridDesc_K0_M_K1,
+          typename BGridDesc_K0_N_K1,
+          typename CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          typename Block2CTileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_wmma(
+            const FloatA* __restrict__ p_a_grid,
+            const FloatB* __restrict__ p_b_grid,
+            FloatC* __restrict__ p_c_grid,
+            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
+            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
+            const CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                c_grid_desc_mblock_mperblock_nblock_nperblock,
+            // const
+            // CGridDescriptor_MBlockxRepeat_MWave_MSubGroup_MAccVgprs_NBlockxRepeat_NWave_NThreadPerSubGroup
+            //     c_grid_desc_mblockxrepeat_mwave_msubgroup_maccvgprs_nblockxrepeat_nwave_nthreadpersubgroup,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CElementwiseOperation c_element_op,
+            const Block2CTileMap block_2_ctile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
+                                                  p_b_grid,
+                                                  p_c_grid,
+                                                  p_shared,
+                                                  a_grid_desc_k0_m_k1,
+                                                  b_grid_desc_k0_n_k1,
+                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  a_element_op,
+                                                  b_element_op,
+                                                  c_element_op,
+                                                  block_2_ctile_map);
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_c_grid;
+    ignore = a_grid_desc_k0_m_k1;
+    ignore = b_grid_desc_k0_n_k1;
+    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = c_element_op;
+    ignore = block_2_ctile_map;
+#endif // end of if (defined(__gfx1100__))
+}
+
+template <index_t BlockSize,
+          typename FloatA,
+          typename FloatB,
+          typename FloatAcc,
+          typename FloatCShuffle,
+          typename FloatC,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename AGridDesc_K0_M_K1,
+          typename BGridDesc_K0_N_K1,
+          typename CGridDesc_M_N,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t MPerWmma,
+          index_t NPerWmma,
+          index_t K1Value,
+          index_t MRepeat,
+          index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_K1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_K1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
+          index_t NumGemmKPrefetchStage = 1,
+          LoopScheduler LoopSched       = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer   = PipelineVersion::v1>
+struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto K1 = Number<K1Value>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    using GridwiseGemmPipe = remove_cvref_t<decltype(
+        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
+    {
+        constexpr auto max_lds_align = K1;
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_k0perblock_mperblock_k1 = [&]() {
+            if constexpr(ABlockLdsExtraM)
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1),
+                    make_tuple(Number<MPerBlock + 1>{} * K1, K1, I1));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_aligned(
+                    make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1), max_lds_align);
+            }
+        }();
+
+        return a_block_desc_k0perblock_mperblock_k1;
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_K0PerBlock_NPerBlock_K1()
+    {
+        constexpr auto max_lds_align = K1;
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_k0perblock_nperblock_k1 = [&]() {
+            if constexpr(BBlockLdsExtraN)
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1),
+                    make_tuple(Number<NPerBlock + 1>{} * K1, K1, I1));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_aligned(
+                    make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1), max_lds_align);
+            }
+        }();
+
+        return b_block_desc_k0perblock_nperblock_k1;
+    }
+
+    __host__ __device__ static constexpr auto
+    // *Caution Here repeat is shuffle repeat
+    GetCShuffleBlockDescriptor_MShRepeat_MPerShRepeat_NShRepeat_NPerShRepeat()
+    {
+        constexpr index_t MWave = MPerBlock / (MRepeat * MPerWmma);
+        constexpr index_t NWave = NPerBlock / (NRepeat * NPerWmma);
+
+        constexpr auto c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMRepeatPerShuffle * MWave * MPerWmma>{},
+                           I1,
+                           Number<CShuffleNRepeatPerShuffle * NWave * NPerWmma>{}));
+
+        return c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat;
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_k0perblock_mperblock_k1 =
+            GetABlockDescriptor_K0PerBlock_MPerBlock_K1();
+
+        constexpr auto b_block_desc_k0perblock_nperblock_k1 =
+            GetBBlockDescriptor_K0PerBlock_NPerBlock_K1();
+
+        constexpr auto max_lds_align = K1;
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_k0perblock_mperblock_k1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_k0perblock_nperblock_k1.GetElementSpaceSize(), max_lds_align);
+
+        return (a_block_space_size_aligned * sizeof(FloatA) +
+                b_block_space_size_aligned * sizeof(FloatB));
+    }
+
+    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
+    template <typename Block2CTileMap>
+    __host__ __device__ static constexpr bool
+    CheckValidity(const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
+                  const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
+                  const CGridDesc_M_N& c_grid_desc_m_n,
+                  const Block2CTileMap& block_2_ctile_map)
+    {
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
+                      "wrong! K1 need to be known at compile-time");
+
+        static_assert((MPerBlock % (MPerWmma * MRepeat) == 0) &&
+                          (NPerBlock % (NRepeat * NPerWmma)) == 0,
+                      "Invalid tuning param!");
+
+        const auto M  = a_grid_desc_k0_m_k1.GetLength(I1);
+        const auto N  = b_grid_desc_k0_n_k1.GetLength(I1);
+        const auto K0 = a_grid_desc_k0_m_k1.GetLength(I0);
+
+        if(!(M == c_grid_desc_m_n.GetLength(I0) && N == c_grid_desc_m_n.GetLength(I1) &&
+             K0 == b_grid_desc_k0_n_k1.GetLength(I0) && K1 == a_grid_desc_k0_m_k1.GetLength(I2) &&
+             K1 == b_grid_desc_k0_n_k1.GetLength(I2)))
+            return false;
+
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K0 % K0PerBlock == 0))
+            return false;
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = K0 / K0PerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        if(!block_2_ctile_map.CheckValidity(c_grid_desc_m_n))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        constexpr long_index_t TwoGB = (long_index_t{1} << 31);
+
+        if(!(a_grid_desc_k0_m_k1.GetElementSpaceSize() * sizeof(FloatA) <= TwoGB &&
+             b_grid_desc_k0_n_k1.GetElementSpaceSize() * sizeof(FloatB) <= TwoGB))
+        {
+            return false;
+        }
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / (K0PerBlock * K1);
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        const auto M = c_grid_desc_m_n.GetLength(I0);
+        const auto N = c_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        const auto c_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            c_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return c_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // return block_id to C matrix tile idx (m0, n0) mapping
+    __host__ __device__ static constexpr auto MakeDefaultBlock2CTileMap(
+        const CGridDesc_M_N& c_grid_desc_m_n, index_t /* M01 */, index_t /* N01 */)
+    {
+        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>(
+            c_grid_desc_m_n);
+    }
+
+    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
+        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
+    using DefaultBlock2CTileMap =
+        remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}, 1, 1))>;
+
+    template <bool HasMainKBlockLoop, typename Block2CTileMap = DefaultBlock2CTileMap>
+    __device__ static void Run(const FloatA* __restrict__ p_a_grid,
+                               const FloatB* __restrict__ p_b_grid,
+                               FloatC* __restrict__ p_c_grid,
+                               void* __restrict__ p_shared,
+                               const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
+                               const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
+                               const CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   c_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CElementwiseOperation& c_element_op,
+                               const Block2CTileMap& block_2_ctile_map)
+    {
+        // clang-format off
+/*******************************************************************************/
+// Memory buffer zone.
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_k0_m_k1.GetElementSpaceSize());
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_k0_n_k1.GetElementSpaceSize());
+        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+/*******************************************************************************/
+// BlockIdx.x -> [BlockId.m, BlockId.n]
+        const auto block_work_idx = block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+        if(!block_2_ctile_map.ValidCTileIndex(
+               block_work_idx,
+               make_tuple(c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
+                          c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
+        { return; }
+
+        // Store BlockId into SGPR
+        const index_t m_block_data_idx_on_grid = __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+        const index_t n_block_data_idx_on_grid = __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
+
+/*******************************************************************************/
+// BlockLevel, A/B Matrix ThreadMapping in LDS, As Destinaion of BlockWise_Copy
+        const auto K0 = a_grid_desc_k0_m_k1.GetLength(I0);
+        constexpr auto max_lds_align = K1;
+        constexpr auto a_block_desc_k0perblock_mperblock_k1 = GetABlockDescriptor_K0PerBlock_MPerBlock_K1();
+        constexpr auto b_block_desc_k0perblock_nperblock_k1 = GetBBlockDescriptor_K0PerBlock_NPerBlock_K1();
+        // A matrix blockwise copy
+        auto a_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<        ThisThreadBlock,
+/* typename SrcElementwiseOperation,              */    AElementwiseOperation,
+/* typename DstElementwiseOperation,              */    ck::tensor_operation::element_wise::PassThrough,
+/* InMemoryDataOperationEnum DstInMemOp,          */    InMemoryDataOperationEnum::Set,
+/* typename BlockSliceLengths,                    */    Sequence<K0PerBlock, MPerBlock, K1>,
+/* typename ThreadClusterLengths,                 */    ABlockTransferThreadClusterLengths_K0_M_K1,
+/* typename ThreadClusterArrangeOrder,            */    ABlockTransferThreadClusterArrangeOrder,
+/* typename SrcData,                              */    FloatA,
+/* typename DstData,                              */    FloatA,
+/* typename SrcDesc,                              */    decltype(a_grid_desc_k0_m_k1),
+/* typename DstDesc,                              */    decltype(a_block_desc_k0perblock_mperblock_k1),
+/* typename SrcDimAccessOrder,                    */    ABlockTransferSrcAccessOrder,
+/* typename DstDimAccessOrder,                    */    Sequence<0, 1, 2>,
+/* index_t SrcVectorDim,                          */    ABlockTransferSrcVectorDim,
+/* index_t DstVectorDim,                          */    2,
+/* index_t SrcScalarPerVector,                    */    ABlockTransferSrcScalarPerVector,
+/* index_t DstScalarPerVector,                    */    ABlockTransferDstScalarPerVector_K1,
+/* index_t SrcScalarStrideInVector,               */    1,
+/* index_t DstScalarStrideInVector,               */    1,
+/* bool ThreadTransferSrcResetCoordinateAfterRun, */    AThreadTransferSrcResetCoordinateAfterRun,
+/* bool ThreadTransferDstResetCoordinateAfterRun, */    true>(
+                a_grid_desc_k0_m_k1,
+                make_multi_index(0, m_block_data_idx_on_grid, 0),
+                a_element_op,
+                a_block_desc_k0perblock_mperblock_k1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // B matrix blockwise copy
+        auto b_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<K0PerBlock, NPerBlock, K1>,
+                                                BBlockTransferThreadClusterLengths_K0_N_K1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                FloatB,
+                                                FloatB,
+                                                decltype(b_grid_desc_k0_n_k1),
+                                                decltype(b_block_desc_k0perblock_nperblock_k1),
+                                                BBlockTransferSrcAccessOrder,
+                                                Sequence<0, 1, 2>,
+                                                BBlockTransferSrcVectorDim,
+                                                2,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_K1,
+                                                1,
+                                                1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                true>(
+                b_grid_desc_k0_n_k1,
+                make_multi_index(0, n_block_data_idx_on_grid, 0),
+                b_element_op,
+                b_block_desc_k0perblock_nperblock_k1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+/*******************************************************************************/
+        // GEMM
+        constexpr auto WmmaK = 16;
+        constexpr auto KPack = math::integer_least_multiple(K1, WmmaK);
+
+        auto blockwise_gemm =
+            BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle<BlockSize,
+                                                         FloatA,
+                                                         FloatB,
+                                                         FloatAcc,
+                                                         decltype(a_block_desc_k0perblock_mperblock_k1),
+                                                         decltype(b_block_desc_k0perblock_nperblock_k1),
+                                                         MPerWmma,
+                                                         NPerWmma,
+                                                         MRepeat,
+                                                         NRepeat,
+                                                         KPack>{};
+
+        // Prepare Register for C matrix
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+/*******************************************************************************/
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(a_block_desc_k0perblock_mperblock_k1.GetElementSpaceSize(), max_lds_align);
+        // LDS allocation for A and B: be careful of alignment
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(static_cast<FloatA*>(p_shared), a_block_desc_k0perblock_mperblock_k1.GetElementSpaceSize());
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(static_cast<FloatB*>(p_shared) + a_block_space_size_aligned, b_block_desc_k0perblock_nperblock_k1.GetElementSpaceSize());
+        
+        // Shift Per SUB_K
+        constexpr auto a_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
+
+        // gridwise GEMM pipeline
+        const index_t K0BlockMainLoop = __builtin_amdgcn_readfirstlane(K0 / K0PerBlock);
+        GridwiseGemmPipe::template Run<HasMainKBlockLoop>(a_grid_desc_k0_m_k1,
+                                                          a_block_desc_k0perblock_mperblock_k1,
+                                                          a_blockwise_copy,
+                                                          a_grid_buf,
+                                                          a_block_buf,
+                                                          a_block_slice_copy_step,
+                                                          b_grid_desc_k0_n_k1,
+                                                          b_block_desc_k0perblock_nperblock_k1,
+                                                          b_blockwise_copy,
+                                                          b_grid_buf,
+                                                          b_block_buf,
+                                                          b_block_slice_copy_step,
+                                                          blockwise_gemm,
+                                                          c_thread_buf,
+                                                          K0BlockMainLoop);
+/*******************************************************************************/
+        // write out to C, implement shuffle
+        {
+            constexpr auto c_thread_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs =  
+            blockwise_gemm.GetCThreadDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs();
+
+            // This API Provide All dimension (size) you need
+            constexpr auto c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp =
+                blockwise_gemm.GetCBlockDescriptor_MRepeat_MWave_MSubGroup_NRepeat_NWave_NThreadPerSubGroup_MAccVgprs();
+
+            constexpr auto MWave              = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I1);
+            constexpr auto MSubGroup          = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I2);
+            constexpr auto NWave              = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I4);
+            constexpr auto NThreadPerSubGroup = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I5);
+            constexpr auto MAccVgprs          = c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs_tmp.GetLength(I6);
+
+            // LDS descriptor, shuffle and write out in MRepeat x NRepeat times
+            constexpr auto c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat =
+                GetCShuffleBlockDescriptor_MShRepeat_MPerShRepeat_NShRepeat_NPerShRepeat();
+
+            auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                static_cast<FloatCShuffle*>(p_shared),
+                c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat.GetElementSpaceSize());
+
+            constexpr auto c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs = transform_tensor_descriptor(
+                c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleMRepeatPerShuffle>{}, // MRepeat per shuffle repeat
+                        MWave,                               // MWave
+                        MSubGroup,                           // MSubGroup * MAccVgprs = MPerWmma
+                        MAccVgprs)),
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleNRepeatPerShuffle>{}, // NRepeat per shuffle repeat
+                        NWave,                               // NWave
+                        NThreadPerSubGroup))),               // NThreadPerSubGroup = NPerWmma
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<>{}, Sequence<0, 1, 2, 6>{}, Sequence<>{}, Sequence<3, 4, 5>{}));
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block = blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0);
+
+            const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+            const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_block_to_mrepeat_mwave_msubgroup_maccvgprs_adaptor =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_merge_transform(make_tuple(MRepeat, MWave, MSubGroup, MAccVgprs))),
+                make_tuple(Sequence<0, 1, 2, 3>{}),
+                make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_block_to_nrepeat_nwave_nthreadpersubgroup_adaptor =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_merge_transform(make_tuple(NRepeat, NWave, NThreadPerSubGroup))),
+                make_tuple(Sequence<0, 1, 2>{}),
+                make_tuple(Sequence<0>{}));
+            
+            const auto m_thread_data_on_block_idx = m_thread_data_on_block_to_mrepeat_mwave_msubgroup_maccvgprs_adaptor.CalculateBottomIndex(
+                make_multi_index(m_thread_data_on_block));
+            
+            const auto n_thread_data_on_block_idx = n_thread_data_on_block_to_nrepeat_nwave_nthreadpersubgroup_adaptor.CalculateBottomIndex(
+                make_multi_index(n_thread_data_on_block));
+
+            // shuffle: threadwise copy C from VGPR to LDS
+            auto c_thread_copy_vgpr_to_lds =
+                ThreadwiseTensorSliceTransfer_v1r3<FloatAcc,
+                                                   FloatCShuffle,
+                                                   decltype(c_thread_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs),
+                                                   decltype(c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs),
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Sequence<CShuffleMRepeatPerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            CShuffleNRepeatPerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            MAccVgprs>,
+                                                   Sequence<0, 1, 2, 3, 4, 5, 6>,
+                                                   6,
+                                                   1, // vector write pixel
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>{
+                    c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs,
+                    make_multi_index(0,
+                                     m_thread_data_on_block_idx[I1],
+                                     m_thread_data_on_block_idx[I2],
+                                     0,
+                                     n_thread_data_on_block_idx[I1],
+                                     n_thread_data_on_block_idx[I2],
+                                     m_thread_data_on_block_idx[I3]),
+                    ck::tensor_operation::element_wise::PassThrough{}};
+
+            // shuffle: blockwise copy C from LDS to global
+            auto c_shuffle_block_copy_lds_to_global = ThreadGroupTensorSliceTransfer_v6r1<
+                ThisThreadBlock,            // ThreadGroup
+                CElementwiseOperation,      // ElementwiseOperation,
+                CGlobalMemoryDataOperation, // DstInMemOp,
+                Sequence<1,
+                         CShuffleMRepeatPerShuffle * MWave * MPerWmma,
+                         1,
+                         CShuffleNRepeatPerShuffle * NWave * NPerWmma>, // BlockSliceLengths,
+                CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                FloatCShuffle,        // typename SrcData,
+                FloatC,               // typename DstData,
+                decltype(c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat),
+                decltype(c_grid_desc_mblock_mperblock_nblock_nperblock),
+                Sequence<0, 1, 2, 3>,                           // typename DimAccessOrder,
+                3,                                              // index_t VectorDim,
+                CShuffleBlockTransferScalarPerVector_NPerBlock, // index_t ScalarPerVector,
+                true,  // bool ThreadTransferSrcResetCoordinateAfterRun,
+                false> // bool ThreadTransferDstResetCoordinateAfterRun>
+                {c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat,
+                 make_multi_index(0, 0, 0, 0),
+                 c_grid_desc_mblock_mperblock_nblock_nperblock,
+                 make_multi_index(block_work_idx[I0], 0, block_work_idx[I1], 0),
+                 c_element_op};
+
+            // space filling curve for local reg & global memory
+            // space filling curve for threadwise C in VGPR
+            constexpr auto sfc_c_vgpr =
+                SpaceFillingCurve<Sequence<MRepeat, 1, 1, NRepeat, 1, 1, MAccVgprs>,
+                                  Sequence<0, 1, 2, 3, 4, 5, 6>,
+                                  Sequence<CShuffleMRepeatPerShuffle,
+                                           1,
+                                           1,
+                                           CShuffleNRepeatPerShuffle,
+                                           1,
+                                           1,
+                                           MAccVgprs>>{};
+
+            // space filling curve for shuffled blockwise C in global mem
+            constexpr auto sfc_c_global =
+                SpaceFillingCurve<Sequence<1, MPerBlock, 1, NPerBlock>,
+                                  Sequence<0, 2, 1, 3>,
+                                  Sequence<1,
+                                           CShuffleMRepeatPerShuffle * MWave * MPerWmma,
+                                           1,
+                                           CShuffleNRepeatPerShuffle * NWave * NPerWmma>>{};
+
+            constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+            static_assert(num_access == sfc_c_global.GetNumOfAccess(), "wrong!");
+
+            static_for<0, num_access, 1>{}([&](auto access_id) {
+                // make sure it's safe to write to LDS
+                block_sync_lds();
+
+                // each thread write its data from VGPR to LDS
+                c_thread_copy_vgpr_to_lds.Run(c_thread_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs,
+                                              sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                              c_thread_buf,
+                                              c_block_desc_mrepeat_mwave_msubgroup_nrepeat_nwave_nthreadpersubgroup_maccvgprs,
+                                              c_shuffle_block_buf);
+
+                // make sure it's safe to read from LDS
+                block_sync_lds();
+
+                // each block copy its data from LDS to global
+                c_shuffle_block_copy_lds_to_global.Run(
+                    c_shuffle_block_desc_mshrepeat_mpershrepeat_nshrepeat_npershrepeat,
+                    c_shuffle_block_buf,
+                    c_grid_desc_mblock_mperblock_nblock_nperblock,
+                    c_grid_buf);
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    constexpr auto c_global_step = sfc_c_global.GetForwardStep(access_id);
+                    // move on C
+                    c_shuffle_block_copy_lds_to_global.MoveDstSliceWindow(
+                        c_grid_desc_mblock_mperblock_nblock_nperblock, c_global_step);
+                }
+            });
+        }
+        // clang-format on
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index ecc528a7ed79a337efd02cc5efe9dd11039b3b29..7f1fb21d33efe67f4bebc8660249b755b9e51914 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,17 +17,25 @@
 
 namespace ck {
 
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename AGridDesc_AK0_M_AK1,
-          typename BGridDesc_BK0_N_BK1,
-          typename CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename Block2CTileMap,
-          bool HasMainKBlockLoop>
+template <typename GridwiseGemm, bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_xdl_cshuffle_v1(typename GridwiseGemm::Argument karg)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(
+        karg.p_a_grid, karg.p_b_grid, karg.p_c_grid, p_shared, karg);
+#else
+    ignore = karg;
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+template <typename GridwiseGemm, typename FloatAB, typename FloatC, bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
@@ -35,54 +43,33 @@ __global__ void
         kernel_gemm_xdl_cshuffle_v1(const FloatAB* __restrict__ p_a_grid,
                                     const FloatAB* __restrict__ p_b_grid,
                                     FloatC* __restrict__ p_c_grid,
-                                    const AElementwiseOperation a_element_op,
-                                    const BElementwiseOperation b_element_op,
-                                    const CElementwiseOperation c_element_op,
-                                    const AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1,
-                                    const BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1,
-                                    const CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                                        c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                    const Block2CTileMap block_2_ctile_map)
+                                    typename GridwiseGemm::Problem problem)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
-                                                  p_b_grid,
-                                                  p_c_grid,
-                                                  p_shared,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  a_grid_desc_ak0_m_ak1,
-                                                  b_grid_desc_bk0_n_bk1,
-                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  block_2_ctile_map);
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid, p_b_grid, p_c_grid, p_shared, problem);
 #else
     ignore = p_a_grid;
     ignore = p_b_grid;
     ignore = p_c_grid;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = a_grid_desc_ak0_m_ak1;
-    ignore = b_grid_desc_bk0_n_bk1;
-    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = block_2_ctile_map;
+    ignore = problem;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
-template <typename FloatAB,
+template <typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename FloatAB,
           typename FloatGemmAcc,
           typename FloatCShuffle,
           typename FloatC,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
+          tensor_operation::device::GemmSpecialization GemmSpec,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_AK0_M_AK1,
-          typename BGridDesc_BK0_N_BK1,
-          typename CGridDesc_M_N,
           index_t NumGemmKPrefetchStage,
           index_t BlockSize,
           index_t MPerBlock,
@@ -128,35 +115,396 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     static constexpr auto I7 = Number<7>{};
 
     // K1 should be Number<...>
-    static constexpr auto AK0 = Number<KPerBlock / AK1Value>{};
-    static constexpr auto BK0 = Number<KPerBlock / BK1Value>{};
-    static constexpr auto AK1 = Number<AK1Value>{};
-    static constexpr auto BK1 = Number<BK1Value>{};
+    static constexpr auto AK0Number = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0Number = Number<KPerBlock / BK1Value>{};
+    static constexpr auto AK1Number = Number<AK1Value>{};
+    static constexpr auto BK1Number = Number<BK1Value>{};
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    __host__ static auto CalculateGridSize(index_t M, index_t N)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
+    }
+
+    __host__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_divide_ceil(M, MPerBlock) * MPerBlock;
+    }
+
+    __host__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_divide_ceil(N, NPerBlock) * NPerBlock;
+    }
+
+    __host__ static auto CalculateKPadded(index_t K)
+    {
+        return math::integer_divide_ceil(K, KPerBlock) * KPerBlock;
+    }
+
+    __host__ static auto CalculateAK0(index_t K)
+    {
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding ||
+                     GemmSpec == GemmSpecialization::KPadding ||
+                     GemmSpec == GemmSpecialization::NKPadding)
+        {
+            return CalculateKPadded(K) / AK1Value;
+        }
+        else
+        {
+            return K / AK1Value;
+        }
+    }
+
+    __host__ static auto CalculateBK0(index_t K)
+    {
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding ||
+                     GemmSpec == GemmSpecialization::KPadding ||
+                     GemmSpec == GemmSpecialization::MKPadding)
+        {
+            return CalculateKPadded(K) / BK1Value;
+        }
+        else
+        {
+            return K / BK1Value;
+        }
+    }
+
+    __host__ static auto CalculateMBlock(index_t M)
+    {
+        return math::integer_divide_floor(M, MPerBlock);
+    }
+
+    __host__ static auto CalculateNBlock(index_t N)
+    {
+        return math::integer_divide_floor(N, NPerBlock);
+    }
+
+    __device__ static auto MakeAGridDescriptor_AK0_M_AK1(
+        index_t M, index_t MPad, index_t K, index_t KPad, index_t StrideA, index_t AK0)
+    {
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding)
+        {
+            // pad both M and K
+            const auto a_grid_desc_m_k =
+                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
+                                            make_tuple(make_right_pad_transform(M, MPad - M),
+                                                       make_right_pad_transform(K, KPad - K)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_pass_through_transform(MPad)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
+                          GemmSpec == GemmSpecialization::MNPadding)
+        {
+            // pad M, but not K
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_mraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
+                          GemmSpec == GemmSpecialization::NKPadding)
+        {
+            // pad K, but not M
+            const auto a_grid_desc_m_k = transform_tensor_descriptor(
+                a_grid_desc_mraw_kraw,
+                make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+        else
+        {
+            // not pad M or K
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_mraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+    }
+
+    __device__ static auto MakeBGridDescriptor_BK0_N_BK1(
+        index_t K, index_t KPad, index_t N, index_t NPad, index_t StrideB, index_t BK0)
+    {
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N, K), make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N, K), make_tuple(StrideB, I1));
+            }
+        }();
+
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding)
+        {
+            // pad both N and K
+            const auto b_grid_desc_n_k =
+                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
+                                            make_tuple(make_right_pad_transform(N, NPad - N),
+                                                       make_right_pad_transform(K, KPad - K)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_n_k,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_pass_through_transform(NPad)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
+                          GemmSpec == GemmSpecialization::MNPadding)
+        {
+            // pad N, but not K
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_nraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
+                          GemmSpec == GemmSpecialization::MKPadding)
+        {
+            // pad K, but not N
+            const auto b_grid_desc_n_k = transform_tensor_descriptor(
+                b_grid_desc_nraw_kraw,
+                make_tuple(make_pass_through_transform(N), make_right_pad_transform(K, KPad - K)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_n_k,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+        else
+        {
+            // not pad N or K
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_nraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+    }
+
+    __host__ __device__ static auto
+    MakeCGridDescriptor_M_N(index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideC)
+    {
+        const auto c_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding)
+        {
+            // pad M and N
+            return transform_tensor_descriptor(c_grid_desc_mraw_nraw,
+                                               make_tuple(make_right_pad_transform(M, MPad - M),
+                                                          make_right_pad_transform(N, NPad - N)),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
+                          GemmSpec == GemmSpecialization::MKPadding)
+        {
+            // pad M, but not N
+            return transform_tensor_descriptor(
+                c_grid_desc_mraw_nraw,
+                make_tuple(make_right_pad_transform(M, MPad - M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
+                          GemmSpec == GemmSpecialization::NKPadding)
+        {
+            // pad N, but not M
+            return transform_tensor_descriptor(
+                c_grid_desc_mraw_nraw,
+                make_tuple(make_pass_through_transform(M), make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+            // not pad M or N
+            return c_grid_desc_mraw_nraw;
+        }
+    }
+
+    struct Problem
+    {
+        __host__ Problem(index_t M_,
+                         index_t N_,
+                         index_t K_,
+                         index_t StrideA_,
+                         index_t StrideB_,
+                         index_t StrideC_)
+            : M{M_},
+              N{N_},
+              K{K_},
+              StrideA{StrideA_},
+              StrideB{StrideB_},
+              StrideC{StrideC_},
+              MPadded{CalculateMPadded(M_)},
+              NPadded{CalculateNPadded(N_)},
+              KPadded{CalculateKPadded(K_)},
+              AK0{CalculateAK0(K_)},
+              BK0{CalculateBK0(K_)},
+              MBlock{CalculateMBlock(M_)},
+              NBlock{CalculateNBlock(N_)}
+        {
+        }
+
+        __host__ void Print() const
+        {
+            std::cout << "problem {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << ", "
+                      << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", "
+                      << "KP:" << KPadded << ", "
+                      << "AK0:" << AK0 << ", "
+                      << "BK0:" << BK0 << ", "
+                      << "MBlock: " << MBlock << ", "
+                      << "NBlock: " << NBlock << "}" << std::endl;
+        }
+
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        index_t MPadded;
+        index_t NPadded;
+        index_t KPadded;
+        index_t AK0;
+        index_t BK0;
+        index_t MBlock;
+        index_t NBlock;
+    };
+
+    // Argument
+    struct Argument : public tensor_operation::device::BaseArgument, public Problem
+    {
+        __host__ Argument(const FloatAB* p_a_grid_,
+                          const FloatAB* p_b_grid_,
+                          FloatC* p_c_grid_,
+                          index_t M_,
+                          index_t N_,
+                          index_t K_,
+                          index_t StrideA_,
+                          index_t StrideB_,
+                          index_t StrideC_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid{p_a_grid_},
+              p_b_grid{p_b_grid_},
+              p_c_grid{p_c_grid_}
+        {
+        }
+
+        const FloatAB* p_a_grid;
+        const FloatAB* p_b_grid;
+        FloatC* p_c_grid;
+    };
+
     // FIXME: pass GridwiseGemmPipe as a template arguement into GridwiseGemm
     using GridwiseGemmPipe = remove_cvref_t<decltype(
         GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
-    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
         // A matrix in LDS memory, dst of blockwise copy
         return make_naive_tensor_descriptor(
-            make_tuple(AK0, Number<MPerBlock>{}, AK1),
-            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+            make_tuple(AK0Number, Number<MPerBlock>{}, AK1Number),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1Number, AK1Number, I1));
     }
 
-    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
     {
         // B matrix in LDS memory, dst of blockwise copy
         return make_naive_tensor_descriptor(
-            make_tuple(BK0, Number<NPerBlock>{}, BK1),
-            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+            make_tuple(BK0Number, Number<NPerBlock>{}, BK1Number),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1Number, BK1Number, I1));
     }
 
-    __host__ __device__ static constexpr auto
-    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    __device__ static constexpr auto GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
     {
         constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
         constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
@@ -171,14 +519,14 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
     }
 
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
     {
         // LDS allocation for A and B: be careful of alignment
         constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
         constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
 
         // lds max alignment
-        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+        constexpr auto max_lds_align = math::lcm(AK1Number, BK1Number);
 
         constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
@@ -199,36 +547,102 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     }
 
     // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
-    template <typename Block2CTileMap>
-    __host__ __device__ static constexpr bool
-    CheckValidity(const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
-                  const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
-                  const CGridDesc_M_N& c_grid_desc_m_n,
-                  const Block2CTileMap& block_2_ctile_map)
+    __host__ static constexpr bool CheckValidity(const Problem& problem)
     {
         static_assert((MPerBlock % (MPerXdl * MXdlPerWave) == 0) &&
                           (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
                       "Invalid tuning param!");
 
-        const auto M = a_grid_desc_ak0_m_ak1.GetLength(I1);
-        const auto N = b_grid_desc_bk0_n_bk1.GetLength(I1);
-        const auto K = a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2);
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.M % MPerBlock == 0))
+            {
+                return false;
+            }
+        }
 
-        if(!(M == c_grid_desc_m_n.GetLength(I0) && N == c_grid_desc_m_n.GetLength(I1)))
-            return false;
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.N % NPerBlock == 0))
+            {
+                return false;
+            }
+        }
 
-        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0))
-            return false;
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::KPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding)
+        {
+            if(!(CalculateKPadded(problem.K) % AK1Value == 0) ||
+               !(CalculateKPadded(problem.K) % BK1Value == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(!(problem.K % AK1Value == 0) || !(problem.K % BK1Value == 0))
+            {
+                return false;
+            }
+        }
 
-        // check gridwise gemm pipeline
-        const auto num_k_loop = K / KPerBlock;
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(problem.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
 
-        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
         {
-            return false;
+            if(problem.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
         }
 
-        if(!block_2_ctile_map.CheckValidity(c_grid_desc_m_n))
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+        {
+            if(problem.N % CShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.M % CShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+            {
+                return false;
+            }
+        }
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = (CalculateAK0(problem.K) * AK1Value) / KPerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
         {
             return false;
         }
@@ -237,22 +651,17 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         return true;
     }
 
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    __host__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
     {
         const index_t num_loop = K / KPerBlock;
 
         return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
 
-    __host__ __device__ static constexpr auto
-    MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc_M_N& c_grid_desc_m_n)
+    template <typename CGridDesc>
+    __device__ static constexpr auto MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+        const CGridDesc& c_grid_desc_m_n, index_t MBlock, index_t NBlock)
     {
-        const auto M = c_grid_desc_m_n.GetLength(I0);
-        const auto N = c_grid_desc_m_n.GetLength(I1);
-
-        const auto MBlock = M / MPerBlock;
-        const auto NBlock = N / NPerBlock;
-
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
             c_grid_desc_m_n,
             make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
@@ -264,33 +673,26 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     }
 
     // return block_id to C matrix tile idx (m0, n0) mapping
-    __host__ __device__ static constexpr auto
-    MakeDefaultBlock2CTileMap(const CGridDesc_M_N& c_grid_desc_m_n)
-    {
-        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>(
-            c_grid_desc_m_n);
-    }
-
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
-
-    using DefaultBlock2CTileMap =
-        remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}))>;
+    using Block2CTileMap = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>;
 
-    template <bool HasMainKBlockLoop, typename Block2CTileMap>
+    template <bool HasMainKBlockLoop>
     __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
                                const FloatAB* __restrict__ p_b_grid,
                                FloatC* __restrict__ p_c_grid,
                                void* __restrict__ p_shared,
-                               const AElementwiseOperation& a_element_op,
-                               const BElementwiseOperation& b_element_op,
-                               const CElementwiseOperation& c_element_op,
-                               const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
-                               const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
-                               const CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
-                                   c_grid_desc_mblock_mperblock_nblock_nperblock,
-                               const Block2CTileMap& block_2_ctile_map)
+                               const Problem& problem)
     {
+        const auto a_grid_desc_ak0_m_ak1 = MakeAGridDescriptor_AK0_M_AK1(
+            problem.M, problem.MPadded, problem.K, problem.KPadded, problem.StrideA, problem.AK0);
+        const auto b_grid_desc_bk0_n_bk1 = MakeBGridDescriptor_BK0_N_BK1(
+            problem.K, problem.KPadded, problem.N, problem.NPadded, problem.StrideB, problem.BK0);
+        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(
+            problem.M, problem.MPadded, problem.N, problem.NPadded, problem.StrideC);
+
+        const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                c_grid_desc_m_n, problem.MBlock, problem.NBlock);
+
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_a_grid, a_grid_desc_ak0_m_ak1.GetElementSpaceSize());
         const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
@@ -298,7 +700,13 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
 
+        const AElementwiseOperation a_element_op{};
+        const BElementwiseOperation b_element_op{};
+        const CElementwiseOperation c_element_op{};
+
         // divide block work by [M, N]
+        const auto block_2_ctile_map = Block2CTileMap{problem.M, problem.N};
+
         const auto block_work_idx =
             block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
 
@@ -318,7 +726,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
 
         // lds max alignment
-        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+        constexpr auto max_lds_align = math::lcm(AK1Number, BK1Number);
 
         // A matrix in LDS memory, dst of blockwise copy
         constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
@@ -332,7 +740,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 AElementwiseOperation,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 InMemoryDataOperationEnum::Set,
-                                                Sequence<AK0, MPerBlock, AK1>,
+                                                Sequence<AK0Number, MPerBlock, AK1Number>,
                                                 ABlockTransferThreadClusterLengths_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
@@ -363,7 +771,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 BElementwiseOperation,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 InMemoryDataOperationEnum::Set,
-                                                Sequence<BK0, NPerBlock, BK1>,
+                                                Sequence<BK0Number, NPerBlock, BK1Number>,
                                                 BBlockTransferThreadClusterLengths_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
@@ -395,8 +803,9 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
-        constexpr index_t KPack = math::max(
-            math::lcm(AK1, BK1), MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+        constexpr index_t KPack =
+            math::max(math::lcm(AK1Number, BK1Number),
+                      MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
@@ -424,8 +833,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             static_cast<FloatAB*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
-        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
-        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1, 0, 0);
+        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1Number, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1Number, 0, 0);
 
         // gridwise GEMM pipeline
         static_assert(std::is_default_constructible_v<GridwiseGemmPipe>);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
index 94e181cd45470c08f4f7b35b84ff991a56babb82..8675350094b385824d39d75d19ce74ec41ef2778 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -57,7 +57,8 @@ __global__ void
             const C0GridDescriptor_NBlock_NPerBlock c0_grid_desc_nblock_nperblock,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     // TODO ANT: separate into MMA + Epilogue
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..31c59d14eddb6983c43ff97b9999f78aa7f980b2
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -0,0 +1,744 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename ABDataType,
+          typename FloatGemmAcc,
+          typename EDataTypeShuffle,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename EElementwiseOperation,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename AGridDesc_M_K,
+          typename BGridDesc_N_K,
+          typename EGridDesc_M_N,
+          index_t NumGemmKPrefetchStage,
+          index_t TileLoadThreadGroupSize,
+          index_t TileMathThreadGroupSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1Value,
+          index_t BK1Value,
+          index_t MPerXdl,
+          index_t NPerXdl,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
+struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_waveletmodel_cshuffle
+{
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto AK1         = Number<AK1Value>{};
+    static constexpr auto BK1         = Number<BK1Value>{};
+    static constexpr auto AK0PerBlock = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0PerBlock = Number<KPerBlock / BK1Value>{};
+
+    struct TileLoadThreadGroup
+    {
+        __device__ static constexpr index_t GetNumOfThread() { return TileLoadThreadGroupSize; }
+
+        __device__ static constexpr bool IsBelong()
+        {
+            return (get_thread_local_1d_id() >= TileLoadThreadGroupSize);
+        }
+
+        __device__ static index_t GetThreadId()
+        {
+            return get_thread_local_1d_id() - TileMathThreadGroupSize;
+        }
+    };
+
+    struct TileMathThreadGroup
+    {
+        __device__ static constexpr index_t GetNumOfThread() { return TileMathThreadGroupSize; }
+
+        __device__ static constexpr bool IsBelong()
+        {
+            return get_thread_local_1d_id() < TileMathThreadGroupSize;
+        }
+
+        __device__ static index_t GetThreadId() { return get_thread_local_1d_id(); }
+    };
+
+    using CShuffleBlockTransferThreadGroup = ThisThreadBlock<TileMathThreadGroupSize>;
+
+    // load and math+store Wave pipelines.
+    // TODO: build pipelines blocks scheduling parallel tasks
+    using GridwiseGemmLoad = GridwiseGemmLoadWave<TileLoadThreadGroup, NumGemmKPrefetchStage>;
+    using GridwiseGemmMath = GridwiseGemmMathWave<TileMathThreadGroup, NumGemmKPrefetchStage>;
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(AK0PerBlock, Number<MPerBlock>{}, AK1),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(BK0PerBlock, Number<NPerBlock>{}, BK1),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    {
+        constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+        constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl>{},
+                           I1,
+                           Number<CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>{}));
+
+        return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+
+        // LDS allocation for C shuffle in LDS
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+        constexpr auto c_block_size =
+            c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
+
+        return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
+                             sizeof(ABDataType),
+                         c_block_size * sizeof(EDataTypeShuffle));
+    }
+
+    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
+    template <typename Block2ETileMap>
+    __host__ __device__ static constexpr bool
+    CheckValidity(const AGridDesc_M_K& a_grid_desc_m_k,
+                  const BGridDesc_N_K& b_grid_desc_n_k,
+                  const EGridDesc_M_N& e_grid_desc_m_n,
+                  const Block2ETileMap& /*block_2_etile_map*/)
+    {
+        static_assert((MPerBlock % (MPerXdl * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
+                      "Invalid tuning param!");
+
+        const auto M = a_grid_desc_m_k.GetLength(I0);
+        const auto N = b_grid_desc_n_k.GetLength(I0);
+        const auto K = a_grid_desc_m_k.GetLength(I1);
+
+        // check consistency of desc
+        if(!(M == e_grid_desc_m_n.GetLength(I0) && N == e_grid_desc_m_n.GetLength(I1) &&
+             K == b_grid_desc_n_k.GetLength(I1)))
+        {
+            return false;
+        }
+
+        // check tile size
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0))
+        {
+            return false;
+        }
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = K / KPerBlock;
+
+        if(!GridwiseGemmMath::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+
+        // check tensor size: cannot be larger than 2GB each
+        constexpr long_index_t TwoGB = (long_index_t{1} << 31);
+
+        if(!(a_grid_desc_m_k.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+             b_grid_desc_n_k.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+             e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EDataType) <= TwoGB))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / KPerBlock;
+
+        return GridwiseGemmMath::CalculateHasMainLoop(num_loop);
+    }
+
+    // return block_id to E matrix tile idx (m0, n0) mapping
+    __host__ __device__ static constexpr auto
+    MakeDefaultBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        constexpr auto M1 = Number<MPerBlock>{};
+        constexpr auto N1 = Number<NPerBlock>{};
+
+        const auto M0 = M / M1;
+        const auto N0 = N / N1;
+
+        constexpr auto M01 = I1;
+        constexpr auto N01 = I1;
+
+        const auto m00_m01_n00_n01_to_m0_n0_block_cluster_adaptor =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_unmerge_transform(make_tuple(M0, M01)),
+                           make_unmerge_transform(make_tuple(N0, N01))),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1, 3>{}));
+
+        const auto cblockid_to_m00_m01_n00_n01_block_cluster_adaptor =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_merge_transform(make_tuple(M0, N0, M01, N01))),
+                make_tuple(Sequence<0, 1, 2, 3>{}),
+                make_tuple(Sequence<0>{}));
+
+        const auto cblockid_to_m0_n0_block_cluster_adaptor =
+            chain_tensor_adaptors(m00_m01_n00_n01_to_m0_n0_block_cluster_adaptor,
+                                  cblockid_to_m00_m01_n00_n01_block_cluster_adaptor);
+
+        return cblockid_to_m0_n0_block_cluster_adaptor;
+    }
+
+    __host__ __device__ static constexpr index_t
+    CalculateGridSize(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        const index_t grid_size = (M / MPerBlock) * (N / NPerBlock);
+
+        return grid_size;
+    }
+
+    // A desc for source in blockwise copy
+    __host__ __device__ static constexpr auto
+    MakeDefaultAGridDescriptor_AK0_M_AK1(const AGridDesc_M_K& a_grid_desc_m_k)
+    {
+        const auto M = a_grid_desc_m_k.GetLength(I0);
+        const auto K = a_grid_desc_m_k.GetLength(I1);
+
+        const auto AK0 = K / AK1;
+
+        return transform_tensor_descriptor(a_grid_desc_m_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
+                                                      make_pass_through_transform(M)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    // B desc for source in blockwise copy
+    __host__ __device__ static constexpr auto
+    MakeDefaultBGridDescriptor_BK0_N_BK1(const BGridDesc_N_K& b_grid_desc_n_k)
+    {
+        const auto N = b_grid_desc_n_k.GetLength(I0);
+        const auto K = b_grid_desc_n_k.GetLength(I1);
+
+        const auto BK0 = K / BK1;
+
+        return transform_tensor_descriptor(b_grid_desc_n_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                                                      make_pass_through_transform(N)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    // E desc for destination in blockwise copy
+    template <typename EGridDescriptor_M_N>
+    __host__ __device__ static constexpr auto MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+        const EGridDescriptor_M_N& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            e_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return e_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
+        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+
+    using DefaultBlock2ETileMap =
+        remove_cvref_t<decltype(MakeDefaultBlock2ETileMap(EGridDesc_M_N{}))>;
+
+    template <bool HasMainKBlockLoop,
+              typename AGridDesc_AK0_M_AK1,
+              typename BGridDesc_BK0_N_BK1,
+              typename Block2ETileMap>
+    __device__ static void Run(const ABDataType* __restrict__ p_a_grid,
+                               const ABDataType* __restrict__ p_b_grid,
+                               EDataType* __restrict__ p_e_grid,
+                               void* __restrict__ p_shared,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const EElementwiseOperation& e_element_op,
+                               const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
+                               const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
+                               const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   e_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        // build loadWave and MathWave pipelines
+        // loadWave and MathWave synchronized through LDS
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataType*>(p_shared) + a_block_space_size_aligned,
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1, 0, 0);
+
+        const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
+            (a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2)) /
+            KPerBlock);
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            block_2_etile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        // HACK: this force m/n_block_data_idx_on_grid into SGPR
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+
+        const index_t n_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
+
+        if(TileLoadThreadGroup::IsBelong())
+        {
+
+            // LoadWave
+            const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                p_a_grid, a_grid_desc_ak0_m_ak1.GetElementSpaceSize());
+            const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                p_b_grid, b_grid_desc_bk0_n_bk1.GetElementSpaceSize());
+
+            // A matrix blockwise copy
+            auto a_blockwise_copy =
+                ThreadGroupTensorSliceTransfer_v4r1<TileLoadThreadGroup,
+                                                    AElementwiseOperation,
+                                                    ck::tensor_operation::element_wise::PassThrough,
+                                                    InMemoryDataOperationEnum::Set,
+                                                    Sequence<AK0PerBlock, MPerBlock, AK1>,
+                                                    ABlockTransferThreadClusterLengths_AK0_M_AK1,
+                                                    ABlockTransferThreadClusterArrangeOrder,
+                                                    ABDataType,
+                                                    ABDataType,
+                                                    decltype(a_grid_desc_ak0_m_ak1),
+                                                    decltype(a_block_desc_ak0_m_ak1),
+                                                    ABlockTransferSrcAccessOrder,
+                                                    Sequence<1, 0, 2>,
+                                                    ABlockTransferSrcVectorDim,
+                                                    2,
+                                                    ABlockTransferSrcScalarPerVector,
+                                                    ABlockTransferDstScalarPerVector_AK1,
+                                                    1,
+                                                    1,
+                                                    AThreadTransferSrcResetCoordinateAfterRun,
+                                                    true,
+                                                    NumGemmKPrefetchStage>(
+                    a_grid_desc_ak0_m_ak1,
+                    make_multi_index(0, m_block_data_idx_on_grid, 0),
+                    a_element_op,
+                    a_block_desc_ak0_m_ak1,
+                    make_multi_index(0, 0, 0),
+                    ck::tensor_operation::element_wise::PassThrough{});
+
+            // B matrix blockwise copy
+            auto b_blockwise_copy =
+                ThreadGroupTensorSliceTransfer_v4r1<TileLoadThreadGroup,
+                                                    BElementwiseOperation,
+                                                    ck::tensor_operation::element_wise::PassThrough,
+                                                    InMemoryDataOperationEnum::Set,
+                                                    Sequence<BK0PerBlock, NPerBlock, BK1>,
+                                                    BBlockTransferThreadClusterLengths_BK0_N_BK1,
+                                                    BBlockTransferThreadClusterArrangeOrder,
+                                                    ABDataType,
+                                                    ABDataType,
+                                                    decltype(b_grid_desc_bk0_n_bk1),
+                                                    decltype(b_block_desc_bk0_n_bk1),
+                                                    BBlockTransferSrcAccessOrder,
+                                                    Sequence<1, 0, 2>,
+                                                    BBlockTransferSrcVectorDim,
+                                                    2,
+                                                    BBlockTransferSrcScalarPerVector,
+                                                    BBlockTransferDstScalarPerVector_BK1,
+                                                    1,
+                                                    1,
+                                                    BThreadTransferSrcResetCoordinateAfterRun,
+                                                    true,
+                                                    NumGemmKPrefetchStage>(
+                    b_grid_desc_bk0_n_bk1,
+                    make_multi_index(0, n_block_data_idx_on_grid, 0),
+                    b_element_op,
+                    b_block_desc_bk0_n_bk1,
+                    make_multi_index(0, 0, 0),
+                    ck::tensor_operation::element_wise::PassThrough{});
+
+            GridwiseGemmLoad::template RunLoadWavePipeline<HasMainKBlockLoop>(
+                a_grid_desc_ak0_m_ak1,
+                a_block_desc_ak0_m_ak1,
+                a_blockwise_copy,
+                a_grid_buf,
+                a_block_buf,
+                a_block_slice_copy_step,
+                b_grid_desc_bk0_n_bk1,
+                b_block_desc_bk0_n_bk1,
+                b_blockwise_copy,
+                b_grid_buf,
+                b_block_buf,
+                b_block_slice_copy_step,
+                num_k_block_main_loop);
+
+            block_sync_lds();
+            block_sync_lds();
+        }
+        else if(TileMathThreadGroup::IsBelong())
+        {
+            // branch early for math wave
+            constexpr index_t KPack =
+                math::max(math::lcm(AK1, BK1),
+                          MfmaSelector<ABDataType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+
+            auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<
+                TileMathThreadGroupSize,
+                ABDataType,
+                FloatGemmAcc,
+                decltype(a_block_desc_ak0_m_ak1),
+                decltype(b_block_desc_bk0_n_bk1),
+                MPerXdl,
+                NPerXdl,
+                MXdlPerWave,
+                NXdlPerWave,
+                KPack>{};
+
+            auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+            auto c_grid_buf   = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                p_e_grid, e_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+            // TODO re-architect LDS+math stages
+            // Writing data to GMEM: only math wave is doing the work in cshuffle
+            GridwiseGemmMath::template RunMathWavePipeline<HasMainKBlockLoop>(
+                a_block_buf, b_block_buf, blockwise_gemm, c_thread_buf, num_k_block_main_loop);
+
+            // GEMM definition
+            //   c_mtx += transpose(a_mtx) * b_mtx
+            //     a_mtx[K0PerBlock, MPerBlock] is in LDS
+            //     b_mtx[K0PerBlock, NPerBlock] is in LDS
+            //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
+            //       register
+            // sanity check
+
+            // shuffle C and write out
+            {
+                static_assert(MXdlPerWave % CShuffleMXdlPerWavePerShuffle == 0 &&
+                                  NXdlPerWave % CShuffleNXdlPerWavePerShuffle == 0,
+                              "wrong!");
+
+                constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+                constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+                // TODO: hacky, fix it!
+                constexpr auto c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2 =
+                    blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+                // TODO: hacky, fix it!
+                // c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp is only used to get lengths
+                constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp =
+                    blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+                constexpr auto M0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I0);
+                constexpr auto N0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I1);
+                constexpr auto M1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I2);
+                constexpr auto N1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I3);
+                constexpr auto M2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I4);
+                constexpr auto M3 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I5);
+                constexpr auto M4 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I6);
+                constexpr auto N2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I7);
+
+                constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+                    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+                auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                    static_cast<EDataTypeShuffle*>(p_shared),
+                    c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+                constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2 = transform_tensor_descriptor(
+                    c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                    make_tuple(
+                        make_freeze_transform(I0),
+                        make_unmerge_transform(make_tuple(
+                            Number<CShuffleMXdlPerWavePerShuffle>{}, // M0 (MXdlPerWave) per shuffle
+                            M1,                                      // M1 = MWave
+                            M2,                                      // M2 * M3 * M4 = MPerXdl
+                            M3,
+                            M4)),
+                        make_freeze_transform(I0),
+                        make_unmerge_transform(make_tuple(
+                            Number<CShuffleNXdlPerWavePerShuffle>{}, // N0 (NXdlPerWave) per shuffle
+                            N1,                                      // N1 = NWave
+                            N2))),                                   // N2 = NPerXdl
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<>{},
+                               Sequence<0, 2, 4, 5, 6>{},
+                               Sequence<>{},
+                               Sequence<1, 3, 7>{}));
+
+                // calculate origin of thread output tensor on global memory
+                // blockwise GEMM c matrix starting index
+                const auto c_thread_mtx_on_block =
+                    blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
+
+                const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+                const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+                const auto m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor =
+                    make_single_stage_tensor_adaptor(
+                        make_tuple(make_merge_transform(make_tuple(M0, M1, M2, M3, M4))),
+                        make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                        make_tuple(Sequence<0>{}));
+
+                const auto m_thread_data_on_block_idx =
+                    m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor.CalculateBottomIndex(
+                        make_multi_index(m_thread_data_on_block));
+
+                const auto n_thread_data_on_block_to_n0_n1_n2_adaptor =
+                    make_single_stage_tensor_adaptor(
+                        make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
+                        make_tuple(Sequence<0, 1, 2>{}),
+                        make_tuple(Sequence<0>{}));
+
+                const auto n_thread_data_on_block_idx =
+                    n_thread_data_on_block_to_n0_n1_n2_adaptor.CalculateBottomIndex(
+                        make_multi_index(n_thread_data_on_block));
+
+                // shuffle: threadwise copy C from VGPR to LDS
+                auto c_thread_copy_vgpr_to_lds = ThreadwiseTensorSliceTransfer_v1r3<
+                    FloatGemmAcc,
+                    EDataTypeShuffle,
+                    decltype(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                    decltype(c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                    ck::tensor_operation::element_wise::PassThrough,
+                    Sequence<CShuffleMXdlPerWavePerShuffle,
+                             CShuffleNXdlPerWavePerShuffle,
+                             I1,
+                             I1,
+                             M2,
+                             I1,
+                             M4,
+                             I1>,
+                    Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                    7,
+                    1,
+                    InMemoryDataOperationEnum::Set,
+                    1,
+                    true>{c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                          make_multi_index(0,
+                                           0,
+                                           m_thread_data_on_block_idx[I1],
+                                           n_thread_data_on_block_idx[I1],
+                                           m_thread_data_on_block_idx[I2],
+                                           m_thread_data_on_block_idx[I3],
+                                           m_thread_data_on_block_idx[I4],
+                                           n_thread_data_on_block_idx[I2]),
+                          ck::tensor_operation::element_wise::PassThrough{}};
+
+                // shuffle: blockwise copy C from LDS to global
+                auto c_shuffle_block_copy_lds_to_global = ThreadGroupTensorSliceTransfer_v6r1<
+                    CShuffleBlockTransferThreadGroup, // ThreadGroup
+                    EElementwiseOperation,            // ElementwiseOperation,
+                    CGlobalMemoryDataOperation,       // DstInMemOp,
+                    Sequence<1,
+                             CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                             1,
+                             CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>, // BlockSliceLengths,
+                    CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                    Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                    EDataTypeShuffle,     // typename SrcData,
+                    EDataType,            // typename DstData,
+                    decltype(c_shuffle_block_desc_mblock_mperblock_nblock_nperblock),
+                    decltype(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                    Sequence<0, 1, 2, 3>,                           // typename DimAccessOrder,
+                    3,                                              // index_t VectorDim,
+                    CShuffleBlockTransferScalarPerVector_NPerBlock, // index_t ScalarPerVector,
+                    true,  // bool ThreadTransferSrcResetCoordinateAfterRun,
+                    false> // bool ThreadTransferDstResetCoordinateAfterRun>
+                    {c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                     make_multi_index(0, 0, 0, 0),
+                     e_grid_desc_mblock_mperblock_nblock_nperblock,
+                     make_multi_index(block_work_idx[I0], 0, block_work_idx[I1], 0),
+                     e_element_op};
+
+                // space filling curve for threadwise C in VGPR
+                constexpr auto sfc_c_vgpr =
+                    SpaceFillingCurve<Sequence<MXdlPerWave, NXdlPerWave, 1, 1, M2, 1, M4, 1>,
+                                      Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                      Sequence<CShuffleMXdlPerWavePerShuffle,
+                                               CShuffleNXdlPerWavePerShuffle,
+                                               1,
+                                               1,
+                                               M2,
+                                               1,
+                                               M4,
+                                               1>>{};
+
+                // space filling curve for shuffled blockwise C in global mem
+                constexpr auto sfc_c_global =
+                    SpaceFillingCurve<Sequence<1, MPerBlock, 1, NPerBlock>,
+                                      Sequence<0, 2, 1, 3>,
+                                      Sequence<1,
+                                               CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                                               1,
+                                               CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>>{};
+
+                constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+                static_assert(num_access == sfc_c_global.GetNumOfAccess(), "wrong!");
+
+                // Different way of getting coalesced writes:
+                // We can get rid of doing cshuffle. Instead of reading A rows in contiguous manner
+                // do it interleaved, then mfma can have nice c-mat layout as below:
+                //
+                // TODO
+                //      We do not need to do LDS swizzle to align global writes writing cache lines:
+                //         v_mfma  cmat, amat, bmat, cmat   - c-mat register layout   are 1xN
+                //                                            elments  (N is vertical or strided
+                //                                            dimension)
+                //         v_mfma  cmat, bmat, amat, cmat   - c-mat register layout   are Mx1
+                //         elments  (M is coalescing
+                //                                            dimension) by enumerating M index in
+                //                                            amat, bmat you can align cmat
+                //                                            register(s) to contiguous M elements
+                //                                            for example
+                //              1st mfma instruction  output space : 0 4 8  12 16 ....
+                //              2nd mfma instruction  output space : 1 5 9  13 17 ....
+                //              3rd mfma instruction  output space : 2 6 10 14 18 ....
+                //              4th mfma instruction  output space : 3 7 11 15 19 ....
+                //              you can pack 4 registers output space into 2WORD and do global write
+                //              (no LDS swizzling required)
+
+                static_for<0, num_access, 1>{}([&](auto access_id) {
+                    // make sure it's safe to write to LDS
+                    block_sync_lds();
+
+                    // each thread write its data from VGPR to LDS
+                    c_thread_copy_vgpr_to_lds.Run(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                                  sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                                  c_thread_buf,
+                                                  c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                                  c_shuffle_block_buf);
+                    // make sure it's safe to read from LDS
+                    block_sync_lds();
+
+                    // each block copy its data from LDS to global
+                    c_shuffle_block_copy_lds_to_global.Run(
+                        c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                        c_shuffle_block_buf,
+                        e_grid_desc_mblock_mperblock_nblock_nperblock,
+                        c_grid_buf);
+
+                    if constexpr(access_id < num_access - 1)
+                    {
+                        constexpr auto c_global_step = sfc_c_global.GetForwardStep(access_id);
+
+                        // move on C
+                        c_shuffle_block_copy_lds_to_global.MoveDstSliceWindow(
+                            e_grid_desc_mblock_mperblock_nblock_nperblock, c_global_step);
+                    }
+                });
+            }
+        }
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
index 126887cbacd727a7c5a3d2026599e0a9c4d9e2e3..0fc18bb923f10e2bd055b0df4ad89be332766620 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -165,16 +165,14 @@ __global__ void
                                       const CElementwiseOperation c_element_op,
                                       const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
                                                   p_b_grid,
                                                   p_c_grid,
-                                                  p_shared_block,
+                                                  p_shared,
                                                   a_b_k0_m_k1_grid_desc,
                                                   b_b_k0_n_k1_grid_desc,
                                                   c_grid_desc_mblock_mperblock_nblock_nperblock,
@@ -183,16 +181,16 @@ __global__ void
                                                   c_element_op,
                                                   c_block_cluster_adaptor);
 #else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_c_grid;
-    ignore = a_b_k0_m_k1_grid_desc;
-    ignore = b_b_k0_n_k1_grid_desc;
-    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = c_block_cluster_adaptor;
+    ignore                = p_a_grid;
+    ignore                = p_b_grid;
+    ignore                = p_c_grid;
+    ignore                = a_b_k0_m_k1_grid_desc;
+    ignore                = b_b_k0_n_k1_grid_desc;
+    ignore                = c_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore                = a_element_op;
+    ignore                = b_element_op;
+    ignore                = c_element_op;
+    ignore                = c_block_cluster_adaptor;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -264,6 +262,16 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
     using GridwiseGemmPipe = remove_cvref_t<decltype(
         GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
+    // denorm test fix, required to work around fp16 mfma issue
+    // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
+    // when mfma if fixed, remove this section and update
+    // FloatABAdjusted -> FloatAB throughout this file
+#if CK_WORKAROUND_DENORM_FIX
+    using FloatABAdjusted = conditional_t<is_same_v<FloatAB, ck::half_t>, ck::bhalf_t, FloatAB>;
+#else
+    using FloatABAdjusted = FloatAB;
+#endif
+
     // M0/M1/M1Padding
     static constexpr auto M1PerBlock = Number<ABlockLdsM1PerBlock>{};
     static constexpr auto M0PerBlock = Number<ABlockLdsM0PerBlock>{};
@@ -605,7 +613,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
     __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
                                const FloatAB* __restrict__ p_b_grid,
                                FloatC* __restrict__ p_c_grid,
-                               FloatAB* __restrict__ p_shared_block,
+                               void* __restrict__ p_shared,
                                const AGridDesc_B_K0_M_K1& a_b_k0_m_k1_grid_desc,
                                const BGridDesc_B_K0_N_K1& b_b_k0_n_k1_grid_desc,
                                const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
@@ -666,7 +674,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
                                                 ABlockTransferThreadClusterLengths_K0_M_K1,
                                                 ABlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
-                                                FloatAB,
+                                                FloatABAdjusted,
                                                 decltype(a_b_k0_m_k1_grid_desc),
                                                 decltype(a_b_k0_m_k1_block_desc),
                                                 ABlockTransferSrcAccessOrder,
@@ -696,7 +704,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
                                                 BBlockTransferThreadClusterLengths_K0_N_K1,
                                                 BBlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
-                                                FloatAB,
+                                                FloatABAdjusted,
                                                 decltype(b_b_k0_n_k1_grid_desc),
                                                 decltype(b_b_k0_n_k1_block_desc),
                                                 BBlockTransferSrcAccessOrder,
@@ -725,11 +733,11 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
         // sanity check
 
         constexpr index_t KPack =
-            math::max(K1, MfmaSelector<FloatAB, MPerXDL, NPerXDL>::selected_mfma.k_per_blk);
+            math::max(K1, MfmaSelector<FloatABAdjusted, MPerXDL, NPerXDL>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatAB,
+                                                                FloatABAdjusted,
                                                                 FloatAcc,
                                                                 decltype(a_k0_m_k1_block_desc),
                                                                 decltype(b_k0_n_k1_block_desc),
@@ -745,16 +753,15 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
         constexpr auto a_block_space_size =
             math::integer_least_multiple(a_k0_m_k1_block_desc.GetElementSpaceSize(), max_lds_align);
 
-        FloatAB* p_a_block = p_shared_block;
-        FloatAB* p_b_block = p_shared_block + a_block_space_size;
-
         constexpr auto a_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0);
         constexpr auto b_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block, a_k0_m_k1_block_desc.GetElementSpaceSize());
+            static_cast<FloatABAdjusted*>(p_shared), a_k0_m_k1_block_desc.GetElementSpaceSize());
+
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block, b_k0_n_k1_block_desc.GetElementSpaceSize());
+            static_cast<FloatABAdjusted*>(p_shared) + a_block_space_size,
+            b_k0_n_k1_block_desc.GetElementSpaceSize());
 
         // gridwise GEMM pipeline
         const index_t K0BlockMainLoop = __builtin_amdgcn_readfirstlane(K0 / K0PerBlock);
@@ -798,8 +805,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
             constexpr auto c_block_desc_mblock_mperblock_nblock_nperblock =
                 GetCBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
 
-            void* p_shared = static_cast<void*>(p_shared_block);
-
             auto c_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
                 static_cast<FloatC*>(p_shared),
                 c_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
index 2aad7128f0681bcd903c2c11249380e3883c3e53..b12bcee0f414c6e56b6550c084630b86ab153bb5 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -44,7 +44,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index d1149c0c2e3c80305e6cb415568cb1fa1fbb7a39..3d1b13f9981f1862a50493c2918371ef69a26ae9 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -7,6 +7,7 @@
 #include "ck/tensor_description/multi_index_transform_helper.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
@@ -21,29 +22,24 @@ template <typename GridwiseGemm,
           typename FloatC,
           typename AGridDesc_K0_M_K1,
           typename BGridDesc_K0_N_K1,
-          typename CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename Block2CTileMap,
+          typename CGridDesc_M_N,
           bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_v2r3(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
-            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
-            const CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-            const AElementwiseOperation a_element_op,
-            const BElementwiseOperation b_element_op,
-            const CElementwiseOperation c_element_op,
-            const Block2CTileMap block_2_ctile_map)
+#if CK_USE_WAVES_PER_EU
+        __attribute__((amdgpu_waves_per_eu(CK_MIN_WAVES_PER_EU, CK_MAX_WAVES_PER_EU)))
+#endif
+        kernel_gemm_xdlops_v2r3(const FloatAB* __restrict__ p_a_grid,
+                                const FloatAB* __restrict__ p_b_grid,
+                                FloatC* __restrict__ p_c_grid,
+                                const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
+                                const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
+                                const CGridDesc_M_N c_grid_desc_m_n)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -52,22 +48,49 @@ __global__ void
                                                   p_shared,
                                                   a_grid_desc_k0_m_k1,
                                                   b_grid_desc_k0_n_k1,
-                                                  c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  block_2_ctile_map);
+                                                  c_grid_desc_m_n);
 #else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_c_grid;
-    ignore = a_grid_desc_k0_m_k1;
-    ignore = b_grid_desc_k0_n_k1;
-    ignore = c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = block_2_ctile_map;
+    ignore                = p_a_grid;
+    ignore                = p_b_grid;
+    ignore                = p_c_grid;
+    ignore                = a_grid_desc_k0_m_k1;
+    ignore                = b_grid_desc_k0_n_k1;
+    ignore                = c_grid_desc_m_n;
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+template <typename GridwiseGemm, bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+#if CK_USE_WAVES_PER_EU
+        __attribute__((amdgpu_waves_per_eu(CK_MIN_WAVES_PER_EU, CK_MAX_WAVES_PER_EU)))
+#endif
+        kernel_gemm_xdlops_v2r3(const typename GridwiseGemm::Argument karg)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    const auto a_grid_desc_k0_m_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeAGridDescriptor_K0_M_K1(
+            karg.M, karg.MPadded, karg.K, karg.K0, karg.StrideA));
+    const auto b_grid_desc_k0_n_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeBGridDescriptor_K0_N_K1(
+            karg.K, karg.N, karg.NPadded, karg.K0, karg.StrideB));
+    const auto c_grid_desc_m_n = amd_wave_read_first_lane(GridwiseGemm::MakeCGridDescriptor_M_N(
+        karg.M, karg.MPadded, karg.N, karg.NPadded, karg.StrideC));
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(karg.p_a_grid,
+                                                  karg.p_b_grid,
+                                                  karg.p_c_grid,
+                                                  p_shared,
+                                                  a_grid_desc_k0_m_k1,
+                                                  b_grid_desc_k0_n_k1,
+                                                  c_grid_desc_m_n);
+#else
+    ignore                = karg;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -76,9 +99,6 @@ template <index_t BlockSize,
           typename FloatAcc,
           typename FloatC,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_K0_M_K1,
-          typename BGridDesc_K0_N_K1,
-          typename CGridDesc_M_N,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
@@ -128,9 +148,118 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    __host__ static auto CalculateGridSize(index_t M, index_t N)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ static auto CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(c_grid_desc_m_n), 1, 1);
+    }
+
+    template <typename>
+    __host__ static auto CalculateGridSize(index_t M, index_t N)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
+    }
+
+    __host__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_divide_ceil(M, MPerBlock) * MPerBlock;
+    }
+
+    __host__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_divide_ceil(N, NPerBlock) * NPerBlock;
+    }
+
+    __host__ static auto CalculateK0(index_t K) { return math::integer_divide_floor(K, K1Value); }
+
+    // Argument
+    struct Problem
+    {
+        __host__ Problem(index_t M_,
+                         index_t N_,
+                         index_t K_,
+                         index_t StrideA_,
+                         index_t StrideB_,
+                         index_t StrideC_)
+            : M{M_},
+              N{N_},
+              K{K_},
+              StrideA{StrideA_},
+              StrideB{StrideB_},
+              StrideC{StrideC_},
+              MPadded{CalculateMPadded(M_)},
+              NPadded{CalculateNPadded(N_)},
+              K0{CalculateK0(K)}
+        {
+        }
+
+        __host__ void Print() const
+        {
+            std::cout << "problem {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << ", "
+                      << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", "
+                      << "K0:" << K0 << "}" << std::endl;
+        }
+
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        index_t MPadded;
+        index_t NPadded;
+        index_t K0;
+    };
+
+    // Argument
+    struct Argument : public Problem, public tensor_operation::device::BaseArgument
+    {
+        __host__ Argument(const FloatAB* p_a_grid_,
+                          const FloatAB* p_b_grid_,
+                          FloatC* p_c_grid_,
+                          index_t M_,
+                          index_t N_,
+                          index_t K_,
+                          index_t StrideA_,
+                          index_t StrideB_,
+                          index_t StrideC_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid{p_a_grid_},
+              p_b_grid{p_b_grid_},
+              p_c_grid{p_c_grid_}
+        {
+        }
+
+        const FloatAB* p_a_grid;
+        const FloatAB* p_b_grid;
+        FloatC* p_c_grid;
+    };
+
     using GridwiseGemmPipe = remove_cvref_t<decltype(
         GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
+    // denorm test fix, required to work around fp16 mfma issue
+    // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
+    // when mfma if fixed, remove this section and update
+    // FloatABAdjusted -> FloatAB throughout this file
+#if CK_WORKAROUND_DENORM_FIX
+    using FloatABAdjusted = conditional_t<is_same_v<FloatAB, ck::half_t>, ck::bhalf_t, FloatAB>;
+#else
+    using FloatABAdjusted = FloatAB;
+#endif
+
     __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
     {
         constexpr auto max_lds_align = K1;
@@ -193,13 +322,11 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         return (a_block_space_size_aligned + b_block_space_size_aligned) * sizeof(FloatAB);
     }
 
-    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
-    template <typename Block2CTileMap>
+    template <typename AGridDesc_K0_M_K1, typename BGridDesc_K0_N_K1, typename CGridDesc_M_N>
     __host__ __device__ static constexpr bool
     CheckValidity(const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
                   const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
-                  const CGridDesc_M_N& c_grid_desc_m_n,
-                  const Block2CTileMap& block_2_ctile_map)
+                  const CGridDesc_M_N& c_grid_desc_m_n)
     {
         static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
                       "wrong! K1 need to be known at compile-time");
@@ -228,7 +355,24 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
             return false;
         }
 
-        if(!block_2_ctile_map.CheckValidity(c_grid_desc_m_n))
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        return true;
+    }
+
+    __host__ static constexpr bool CheckValidity(const Problem& problem)
+    {
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
+                      "wrong! K1 need to be known at compile-time");
+
+        static_assert((MPerBlock % (MPerXDL * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXDL)) == 0,
+                      "Invalid tuning param!");
+
+        // check gridwise gemm pipeline
+        const index_t K0      = problem.K / K1Value;
+        const auto num_k_loop = K0 / K0PerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
         {
             return false;
         }
@@ -237,15 +381,16 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         return true;
     }
 
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    __host__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
     {
         const index_t num_loop = K / (K0PerBlock * K1);
 
         return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
 
+    template <typename CGridDesc>
     __host__ __device__ static constexpr auto
-    MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(const CGridDesc_M_N& c_grid_desc_m_n)
+    MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(const CGridDesc& c_grid_desc_m_n)
     {
         constexpr auto max_lds_align = K1;
 
@@ -281,7 +426,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
 
         using BlockwiseGemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatAB,
+                                                                FloatABAdjusted,
                                                                 FloatAcc,
                                                                 decltype(a_block_desc_k0_m_k1),
                                                                 decltype(b_block_desc_k0_n_k1),
@@ -295,31 +440,23 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
     }
 
     // return block_id to C matrix tile idx (m0, n0) mapping
-    __host__ __device__ static constexpr auto MakeDefaultBlock2CTileMap(
-        const CGridDesc_M_N& c_grid_desc_m_n, index_t /* M01 */, index_t /* N01 */)
+    using Block2CTileMap = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>;
+
+    template <bool HasMainKBlockLoop,
+              typename AGridDesc_K0_M_K1,
+              typename BGridDesc_K0_N_K1,
+              typename CGridDesc_M_N>
+    __device__ static void Run(const FloatAB* p_a_grid,
+                               const FloatAB* p_b_grid,
+                               FloatC* p_c_grid,
+                               void* __restrict__ p_shared,
+                               const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
+                               const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
+                               const CGridDesc_M_N& c_grid_desc_m_n)
     {
-        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>(
-            c_grid_desc_m_n);
-    }
+        const auto c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2 =
+            MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n);
 
-    using CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 =
-        decltype(MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}));
-    using DefaultBlock2CTileMap = decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}, 1, 1));
-
-    template <bool HasMainKBlockLoop, typename Block2CTileMap = DefaultBlock2CTileMap>
-    __device__ static void
-    Run(const FloatAB* __restrict__ p_a_grid,
-        const FloatAB* __restrict__ p_b_grid,
-        FloatC* __restrict__ p_c_grid,
-        void* __restrict__ p_shared,
-        const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
-        const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
-        const CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2& c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-        const AElementwiseOperation& a_element_op,
-        const BElementwiseOperation& b_element_op,
-        const CElementwiseOperation& c_element_op,
-        const Block2CTileMap& block_2_ctile_map)
-    {
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_a_grid, a_grid_desc_k0_m_k1.GetElementSpaceSize());
         const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
@@ -327,7 +464,12 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_c_grid, c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2.GetElementSpaceSize());
 
-        const auto K0 = a_grid_desc_k0_m_k1.GetLength(I0);
+        const AElementwiseOperation a_element_op{};
+        const BElementwiseOperation b_element_op{};
+        const CElementwiseOperation c_element_op{};
+
+        const auto block_2_ctile_map =
+            Block2CTileMap{c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1)};
 
         // divide block work by [M, N]
         const auto block_work_idx =
@@ -367,7 +509,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
                                                 ABlockTransferThreadClusterLengths_K0_M_K1,
                                                 ABlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
-                                                FloatAB,
+                                                FloatABAdjusted,
                                                 decltype(a_grid_desc_k0_m_k1),
                                                 decltype(a_block_desc_k0_m_k1),
                                                 ABlockTransferSrcAccessOrder,
@@ -398,7 +540,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
                                                 BBlockTransferThreadClusterLengths_K0_N_K1,
                                                 BBlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
-                                                FloatAB,
+                                                FloatABAdjusted,
                                                 decltype(b_grid_desc_k0_n_k1),
                                                 decltype(b_block_desc_k0_n_k1),
                                                 BBlockTransferSrcAccessOrder,
@@ -428,7 +570,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         // sanity check
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            FloatAB,
+            FloatABAdjusted,
             FloatAcc,
             decltype(a_block_desc_k0_m_k1),
             decltype(b_block_desc_k0_n_k1),
@@ -446,16 +588,17 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
             math::integer_least_multiple(a_block_desc_k0_m_k1.GetElementSpaceSize(), max_lds_align);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<FloatAB*>(p_shared), a_block_desc_k0_m_k1.GetElementSpaceSize());
+            static_cast<FloatABAdjusted*>(p_shared), a_block_desc_k0_m_k1.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<FloatAB*>(p_shared) + a_block_space_size_aligned,
+            static_cast<FloatABAdjusted*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_k0_n_k1.GetElementSpaceSize());
 
         constexpr auto a_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
         constexpr auto b_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
 
         // gridwise GEMM pipeline
+        const auto K0                       = a_grid_desc_k0_m_k1.GetLength(I0);
         const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(K0 / K0PerBlock);
 
         GridwiseGemmPipe::template Run<HasMainKBlockLoop>(a_grid_desc_k0_m_k1,
@@ -554,4 +697,309 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
     }
 };
 
+template <index_t BlockSize,
+          typename FloatAB,
+          typename FloatAcc,
+          typename FloatC,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          tensor_operation::device::GemmSpecialization GemmSpec,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t MPerXDL,
+          index_t NPerXDL,
+          index_t K1Value,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_K1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_K1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          bool BBlockLdsExtraN,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector,
+          index_t NumGemmKPrefetchStage = 1,
+          LoopScheduler LoopSched       = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer   = PipelineVersion::v1>
+struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext
+    : GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<BlockSize,
+                                              FloatAB,
+                                              FloatAcc,
+                                              FloatC,
+                                              CGlobalMemoryDataOperation,
+                                              AElementwiseOperation,
+                                              BElementwiseOperation,
+                                              CElementwiseOperation,
+                                              MPerBlock,
+                                              NPerBlock,
+                                              K0PerBlock,
+                                              MPerXDL,
+                                              NPerXDL,
+                                              K1Value,
+                                              MXdlPerWave,
+                                              NXdlPerWave,
+                                              ABlockTransferThreadClusterLengths_K0_M_K1,
+                                              ABlockTransferThreadClusterArrangeOrder,
+                                              ABlockTransferSrcAccessOrder,
+                                              ABlockTransferSrcVectorDim,
+                                              ABlockTransferSrcScalarPerVector,
+                                              ABlockTransferDstScalarPerVector_K1,
+                                              AThreadTransferSrcResetCoordinateAfterRun,
+                                              ABlockLdsExtraM,
+                                              BBlockTransferThreadClusterLengths_K0_N_K1,
+                                              BBlockTransferThreadClusterArrangeOrder,
+                                              BBlockTransferSrcAccessOrder,
+                                              BBlockTransferSrcVectorDim,
+                                              BBlockTransferSrcScalarPerVector,
+                                              BBlockTransferDstScalarPerVector_K1,
+                                              BThreadTransferSrcResetCoordinateAfterRun,
+                                              BBlockLdsExtraN,
+                                              CThreadTransferSrcDstAccessOrder,
+                                              CThreadTransferSrcDstVectorDim,
+                                              CThreadTransferDstScalarPerVector,
+                                              NumGemmKPrefetchStage,
+                                              LoopSched,
+                                              PipelineVer>
+{
+    using Parent =
+        GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<BlockSize,
+                                                FloatAB,
+                                                FloatAcc,
+                                                FloatC,
+                                                CGlobalMemoryDataOperation,
+                                                AElementwiseOperation,
+                                                BElementwiseOperation,
+                                                CElementwiseOperation,
+                                                MPerBlock,
+                                                NPerBlock,
+                                                K0PerBlock,
+                                                MPerXDL,
+                                                NPerXDL,
+                                                K1Value,
+                                                MXdlPerWave,
+                                                NXdlPerWave,
+                                                ABlockTransferThreadClusterLengths_K0_M_K1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                ABlockTransferSrcAccessOrder,
+                                                ABlockTransferSrcVectorDim,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_K1,
+                                                AThreadTransferSrcResetCoordinateAfterRun,
+                                                ABlockLdsExtraM,
+                                                BBlockTransferThreadClusterLengths_K0_N_K1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                BBlockTransferSrcAccessOrder,
+                                                BBlockTransferSrcVectorDim,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_K1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                BBlockLdsExtraN,
+                                                CThreadTransferSrcDstAccessOrder,
+                                                CThreadTransferSrcDstVectorDim,
+                                                CThreadTransferDstScalarPerVector,
+                                                NumGemmKPrefetchStage,
+                                                LoopSched,
+                                                PipelineVer>;
+
+    using typename Parent::GridwiseGemmPipe;
+    using typename Parent::Problem;
+
+    using Parent::I1;
+
+    using Parent::K1;
+
+    __device__ static auto
+    MakeAGridDescriptor_K0_M_K1(index_t M, index_t MPad, index_t K, index_t K0, index_t StrideA)
+    {
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    __device__ static auto
+    MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t NPad, index_t K0, index_t StrideB)
+    {
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    __device__ static auto
+    MakeCGridDescriptor_M_N(index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideC)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        {
+            return transform_tensor_descriptor(c_grid_desc_m_n,
+                                               make_tuple(make_right_pad_transform(M, MPad - M),
+                                                          make_right_pad_transform(N, NPad - N)),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+    }
+
+    __host__ static constexpr bool CheckValidity(const Problem& problem)
+    {
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
+                      "wrong! K1 need to be known at compile-time");
+
+        static_assert((MPerBlock % (MPerXDL * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXDL)) == 0,
+                      "Invalid tuning param!");
+
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.M % MPerBlock == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.N % NPerBlock == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(problem.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            if(problem.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        // check gridwise gemm pipeline
+        const index_t K0      = problem.K / K1;
+        const auto num_k_loop = K0 / K0PerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        return true;
+    }
+};
+
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
index 949d56483669946f12f40adca568d4e2a6b0bb5b..19fbee727f77ae8b6b6f90a5a6f520accde95332 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -42,7 +42,8 @@ __global__ void
                                 const CElementwiseOperation c_element_op,
                                 const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 190194f1eb18945eb0e6dc3334e9e6a5a5248267..4cd70cc90288342298937845cca09573a40f5881 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -8,70 +8,39 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
 #include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
 #include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 
 namespace ck {
 
 template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_B_K0_M_K1,
-          typename BGridDesc_B_K0_N_K1,
-          typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename CBlockClusterAdaptor,
-          bool HasMainKBlockLoop>
+          bool HasMainKBlockLoop,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename Block2CTileMap>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_v2r4r2(const FloatAB* __restrict__ p_a_grid,
-                                  const FloatAB* __restrict__ p_b_grid,
-                                  FloatC* __restrict__ p_c_grid,
-                                  const AGridDesc_B_K0_M_K1 a_b_k0_m_k1_grid_desc,
-                                  const BGridDesc_B_K0_N_K1 b_b_k0_n_k1_grid_desc,
-                                  const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
-                                      c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                  const AElementwiseOperation a_element_op,
-                                  const BElementwiseOperation b_element_op,
-                                  const CElementwiseOperation c_element_op,
-                                  const CBlockClusterAdaptor c_block_cluster_adaptor)
+        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg,
+                                             const Block2CTileMap& b2c_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
-                                                  p_b_grid,
-                                                  p_c_grid,
-                                                  static_cast<void*>(p_shared_block),
-                                                  a_b_k0_m_k1_grid_desc,
-                                                  b_b_k0_n_k1_grid_desc,
-                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  c_block_cluster_adaptor);
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
+
+    __shared__ uint8_t p_shared[shared_size];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
+        karg, static_cast<void*>(p_shared), b2c_map);
 #else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_c_grid;
-    ignore = a_b_k0_m_k1_grid_desc;
-    ignore = b_b_k0_n_k1_grid_desc;
-    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = c_block_cluster_adaptor;
+    ignore = karg;
+    ignore = b2c_map;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -79,13 +48,14 @@ template <index_t BlockSize,
           typename FloatAB,
           typename FloatAcc,
           typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_B_K0_M_K1,
-          typename BGridDesc_B_K0_N_K1,
-          typename CMNGridDesc,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
+          tensor_operation::device::GemmSpecialization GemmSpec,
+          index_t NumGemmKPrefetchStage,
           index_t MPerBlock,
           index_t NPerBlock,
           index_t K0PerBlock,
@@ -113,7 +83,9 @@ template <index_t BlockSize,
           index_t CShuffleMRepeatPerShuffle,
           index_t CShuffleNRepeatPerShuffle,
           index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
-          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>
+          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1>
 struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 {
     static constexpr auto I0 = Number<0>{};
@@ -126,10 +98,229 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     static constexpr auto I7 = Number<7>{};
 
     // K1 should be Number<...>
-    static constexpr auto K1 = Number<K1Value>{};
+    static constexpr auto K1  = Number<K1Value>{};
+    static constexpr auto M01 = 1;
+    static constexpr auto N01 = 1;
+
+    static constexpr auto gemm_padder =
+        tensor_operation::device::GemmPadder<GemmSpec, index_t, index_t, index_t>{
+            MPerBlock, NPerBlock, K1* K0PerBlock};
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    using GridwiseGemmPipe = remove_cvref_t<decltype(
+        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
+    struct Argument : public ck::tensor_operation::device::BaseArgument
+    {
+        const FloatAB* p_a_grid;
+        const FloatAB* p_b_grid;
+        FloatC* p_c_grid;
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        index_t MPadded;
+        index_t NPadded;
+        index_t KPadded;
+        index_t K0;
+        index_t k_batch;
+
+        Argument(const FloatAB* p_a_grid_,
+                 const FloatAB* p_b_grid_,
+                 FloatC* p_c_grid_,
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_,
+                 index_t MPadded_,
+                 index_t NPadded_,
+                 index_t KPadded_,
+                 index_t K0_,
+                 index_t k_batch_)
+            : p_a_grid(p_a_grid_),
+              p_b_grid(p_b_grid_),
+              p_c_grid(p_c_grid_),
+              M(M_),
+              N(N_),
+              K(K_),
+              StrideA(StrideA_),
+              StrideB(StrideB_),
+              StrideC(StrideC_),
+              MPadded(MPadded_),
+              NPadded(NPadded_),
+              KPadded(KPadded_),
+              K0(K0_),
+              k_batch(k_batch_)
+        {
+        }
+
+        void Print() const
+        {
+            std::cout << "arg {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << ", "
+                      << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", "
+                      << "KP:" << KPadded << ", "
+                      << "K0:" << K0 << ", "
+                      << "KB:" << k_batch << "}" << std::endl;
+        }
+    };
+
+    __host__ __device__ static auto CalculateGridSize(const Argument& karg)
+    {
+        return std::make_tuple(math::integer_divide_ceil(karg.N, NPerBlock),
+                               math::integer_divide_ceil(karg.M, MPerBlock),
+                               karg.k_batch);
+    }
+
+    // prefer this to be called on host
+    __host__ __device__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_least_multiple(M, MPerBlock);
+    }
+
+    __host__ __device__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_least_multiple(N, NPerBlock);
+    }
+
+    __host__ __device__ static auto CalculateK0(index_t K, index_t K_Batch = 1)
+    {
+        // k_batch * k0 * k0_per_block * k1
+        auto K_t = K_Batch * K0PerBlock * K1;
+        return (K + K_t - 1) / K_t * K0PerBlock;
+    }
+
+    __host__ __device__ static auto CalculateKPadded(index_t K, index_t K_Batch = 1)
+    {
+        auto K0 = CalculateK0(K, K_Batch);
+        return K_Batch * K0 * K1;
+    }
+
+    __host__ __device__ static auto MakeAGridDescriptor_KBatch_K0_M_K1(index_t M,
+                                                                       index_t MPad,
+                                                                       index_t K,
+                                                                       index_t StrideA,
+                                                                       index_t KBatch,
+                                                                       index_t K0,
+                                                                       index_t KPad)
+    {
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    __host__ __device__ static auto MakeBGridDescriptor_KBatch_K0_N_K1(index_t K,
+                                                                       index_t NPad,
+                                                                       index_t N,
+                                                                       index_t StrideB,
+                                                                       index_t KBatch,
+                                                                       index_t K0,
+                                                                       index_t KPad)
+    {
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
+            b_grid_desc_k_n,
+            make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    __host__ __device__ static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        return gemm_padder.PadCDescriptor_M_N(c_grid_desc_m_n);
+    }
+
     __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
     {
         constexpr auto max_lds_align = K1;
@@ -178,54 +369,161 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                          c_block_size * sizeof(FloatC));
     }
 
-    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
-    template <typename Block2CTileMap>
-    __host__ __device__ static constexpr bool
-    CheckValidity(const AGridDesc_B_K0_M_K1& a_b_k0_m_k1_grid_desc,
-                  const BGridDesc_B_K0_N_K1& b_b_k0_n_k1_grid_desc,
-                  const CMNGridDesc& c_m_n_grid_desc,
-                  const Block2CTileMap& block_2_ctile_map)
+    __host__ __device__ static constexpr bool CheckValidity(const Argument& karg)
     {
-        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
-                      "wrong! K1 need to be known at compile-time");
-
-        static_assert((MPerBlock % (MPerXDL * MRepeat) == 0) &&
-                          (NPerBlock % (NRepeat * NPerXDL)) == 0,
-                      "Invalid tuning param!");
-
-        const auto M      = a_b_k0_m_k1_grid_desc.GetLength(I2);
-        const auto N      = b_b_k0_n_k1_grid_desc.GetLength(I2);
-        const auto K0     = a_b_k0_m_k1_grid_desc.GetLength(I1);
-        const auto KBatch = a_b_k0_m_k1_grid_desc.GetLength(I0);
-
-        if(!(M == c_m_n_grid_desc.GetLength(I0) && N == c_m_n_grid_desc.GetLength(I1) &&
-             K0 == b_b_k0_n_k1_grid_desc.GetLength(I1) &&
-             K1 == a_b_k0_m_k1_grid_desc.GetLength(I3) &&
-             K1 == b_b_k0_n_k1_grid_desc.GetLength(I3) &&
-             KBatch == b_b_k0_n_k1_grid_desc.GetLength(I0)))
-            return false;
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(karg.M % MPerBlock == 0))
+            {
+#if DEBUG_LOG
+                std::cout << "Arg M value is not a multiple of MPerBlock! M: " << karg.M << " "
+                          << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
+                          << std::endl;
 
-        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K0 % K0PerBlock == 0))
-            return false;
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(karg.N % NPerBlock == 0))
+            {
+#if DEBUG_LOG
+                std::cout << "Arg N value is not a multiple of NPerBlock! N: " << karg.N << " "
+                          << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
+                          << std::endl;
+
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(karg.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg K (" << karg.K
+                          << ") value is not a multiple of ABlockTransferSrcScalarPerVector ("
+                          << ABlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
+        else
+        {
+            if(karg.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg M (" << karg.M
+                          << ") value is not a multiple of ABlockTransferSrcScalarPerVector ("
+                          << ABlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            if(karg.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg N (" << karg.N
+                          << ") value is not a multiple of BBlockTransferSrcScalarPerVector ("
+                          << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
+        else
+        {
+            if(karg.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg K (" << karg.K
+                          << ") value is not a multiple of BBlockTransferSrcScalarPerVector ("
+                          << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
 
-        if(!block_2_ctile_map.CheckValidity(c_m_n_grid_desc))
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
         {
+            if(karg.N % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+            {
+#if DEBUG_LOG
+                std::cout
+                    << "Arg N (" << karg.N
+                    << ") value is not a multiple of CBlockTransferScalarPerVector_NWaveNPerXDL ("
+                    << CBlockTransferScalarPerVector_NWaveNPerXDL << " )! " << __FILE__ << ":"
+                    << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
+        else
+        {
+            if(karg.M % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+            {
+#if DEBUG_LOG
+                std::cout
+                    << "Arg M (" << karg.M
+                    << ") value is not a multiple of CBlockTransferScalarPerVector_NWaveNPerXDL ("
+                    << CBlockTransferScalarPerVector_NWaveNPerXDL << " )! " << __FILE__ << ":"
+                    << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
+                return false;
+            }
+        }
+
+        const auto num_k_loop = karg.K0 / K0PerBlock;
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+#if DEBUG_LOG
+            std::cout << "The number of k loops (" << num_k_loop
+                      << ") value is not supported by GridwiseGemm Pipeline."
+                      << " K0: " << karg.K0 << ", K0PerBlock: " << K0PerBlock << " " << __FILE__
+                      << ":" << __LINE__ << ", in function: " << __func__ << std::endl;
+#endif // DEBUG_LOG
             return false;
         }
 
-        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
         return true;
     }
 
-    __host__ __device__ static constexpr bool CalculateHasMainK0BlockLoop(index_t K0)
+    __host__ __device__ static auto GetKPad(index_t K, index_t KBatch)
     {
-        const bool has_main_k0_block_loop = K0 > K0PerBlock;
+        const index_t K0   = math::integer_divide_ceil(K, K1 * K0PerBlock * KBatch) * K0PerBlock;
+        const index_t KPad = KBatch * K0 * K1;
+        return KPad;
+    }
 
-        return has_main_k0_block_loop;
+    __host__ __device__ static constexpr bool CalculateHasMainK0BlockLoop(index_t K0)
+    {
+        const index_t num_loop = K0 / K0PerBlock;
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
 
+    template <typename CGridDesc>
     __host__ __device__ static constexpr auto
-    MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(const CMNGridDesc& c_m_n_grid_desc)
+    MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc& c_m_n_grid_desc)
     {
         const auto M = c_m_n_grid_desc.GetLength(I0);
         const auto N = c_m_n_grid_desc.GetLength(I1);
@@ -242,10 +540,11 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     }
 
     // return block_id to C matrix tile idx (m0, n0) mapping
+    template <typename CGridDesc>
     __host__ __device__ static constexpr auto MakeCBlockClusterAdaptor(
-        const CMNGridDesc& c_m_n_grid_desc, index_t /* M01 */, index_t /* N01 */, index_t KBatch)
+        const CGridDesc& c_m_n_grid_desc, index_t /* M01 */, index_t /* N01 */, index_t KBatch)
     {
-        return BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CMNGridDesc>(
+        return BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc>(
             c_m_n_grid_desc, 8, KBatch);
     }
 
@@ -262,24 +561,37 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                        Number<CShuffleNRepeatPerShuffle * NWave * NPerXDL>{}));
     }
 
-    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
-        decltype(MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CMNGridDesc{}));
-    using CBlockClusterAdaptor = decltype(MakeCBlockClusterAdaptor(CMNGridDesc{}, 1, 1, 1));
+    // return block_id to C matrix tile idx (m0, n0, k_split) mapping
+    __host__ __device__ static constexpr auto MakeDefaultBlock2CTileMap()
+    {
+        return BlockToCTileMap_3DGrid_KSplit<MPerBlock, NPerBlock>();
+    }
+
+    using CGridDesc_M_N         = remove_cvref_t<decltype(MakeCGridDescriptor_M_N(1, 1, 1))>;
+    using DefaultBlock2CTileMap = remove_cvref_t<decltype(MakeDefaultBlock2CTileMap())>;
 
-    template <bool HasMainKBlockLoop>
-    __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
-                               const FloatAB* __restrict__ p_b_grid,
-                               FloatC* __restrict__ p_c_grid,
+    template <bool HasMainKBlockLoop,
+              InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+              typename Block2CTileMap>
+    __device__ static void Run(const Argument& karg,
                                void* __restrict__ p_shared_block,
-                               const AGridDesc_B_K0_M_K1& a_b_k0_m_k1_grid_desc,
-                               const BGridDesc_B_K0_N_K1& b_b_k0_n_k1_grid_desc,
-                               const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
-                                   c_grid_desc_mblock_mperblock_nblock_nperblock,
-                               const AElementwiseOperation& a_element_op,
-                               const BElementwiseOperation& b_element_op,
-                               const CElementwiseOperation& c_element_op,
-                               const CBlockClusterAdaptor& c_block_cluster_adaptor)
+                               const Block2CTileMap& block_2_ctile_map)
     {
+        const FloatAB* p_a_grid          = karg.p_a_grid;
+        const FloatAB* p_b_grid          = karg.p_b_grid;
+        FloatC* p_c_grid                 = karg.p_c_grid;
+        const auto a_b_k0_m_k1_grid_desc = MakeAGridDescriptor_KBatch_K0_M_K1(
+            karg.M, karg.MPadded, karg.K, karg.StrideA, karg.k_batch, karg.K0, karg.KPadded);
+        const auto b_b_k0_n_k1_grid_desc = MakeBGridDescriptor_KBatch_K0_N_K1(
+            karg.K, karg.NPadded, karg.N, karg.StrideB, karg.k_batch, karg.K0, karg.KPadded);
+        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
+
+        const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
+        const AElementwiseOperation a_element_op = AElementwiseOperation{};
+        const BElementwiseOperation b_element_op = BElementwiseOperation{};
+        const CElementwiseOperation c_element_op = CElementwiseOperation{};
+
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_a_grid, a_b_k0_m_k1_grid_desc.GetElementSpaceSize());
         const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
@@ -287,28 +599,28 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
 
-        const auto K0 = a_b_k0_m_k1_grid_desc.GetLength(I1);
-
-        // divide block work by [M, N]
+        // divide block work by [KBatch, M, N]
         const auto block_work_idx =
-            c_block_cluster_adaptor.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+            block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
 
-        if(!c_block_cluster_adaptor.ValidCTileIndex(
-               make_tuple(block_work_idx[I1], block_work_idx[I2]),
+        if(!block_2_ctile_map.ValidCTileIndex(
+               block_work_idx,
                make_tuple(c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
                           c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
         {
             return;
         }
 
-        const index_t k_batch_id = block_work_idx[I0];
+        const index_t block_m_id = __builtin_amdgcn_readfirstlane(block_work_idx[I1]);
+        const index_t block_n_id = __builtin_amdgcn_readfirstlane(block_work_idx[I2]);
+        const index_t k_batch_id = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
 
         // HACK: this force m/n_block_data_idx_on_grid into SGPR
         const index_t m_block_data_idx_on_grid =
-            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * MPerBlock);
+            __builtin_amdgcn_readfirstlane(block_m_id * MPerBlock);
 
         const index_t n_block_data_idx_on_grid =
-            __builtin_amdgcn_readfirstlane(block_work_idx[I2] * NPerBlock);
+            __builtin_amdgcn_readfirstlane(block_n_id * NPerBlock);
 
         // lds max alignment
         constexpr auto max_lds_align = K1;
@@ -445,17 +757,18 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         //       register
         // sanity check
 
-        auto blockwise_gemm =
-            BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatAB,
-                                                                FloatAcc,
-                                                                decltype(a_k0_m_k1_block_desc),
-                                                                decltype(b_k0_n_k1_block_desc),
-                                                                MPerXDL,
-                                                                NPerXDL,
-                                                                MRepeat,
-                                                                NRepeat,
-                                                                K1>{};
+        auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
+            BlockSize,
+            FloatAB,
+            FloatAcc,
+            decltype(a_k0_m_k1_block_desc),
+            decltype(b_k0_n_k1_block_desc),
+            MPerXDL,
+            NPerXDL,
+            MRepeat,
+            NRepeat,
+            K1,
+            LoopSched>();
 
         auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
 
@@ -474,6 +787,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
             p_b_block, b_k0_n_k1_block_desc.GetElementSpaceSize());
 
+#if 0
         // preload data into LDS
         {
             a_blockwise_copy.RunRead(a_b_k0_m_k1_grid_desc, a_grid_buf);
@@ -510,7 +824,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                 b_blockwise_copy.RunWrite(b_b_k0_n_k1_block_desc, b_block_buf);
 
                 k0_block_data_begin += K0PerBlock;
-            } while(k0_block_data_begin < (K0 - K0PerBlock));
+            } while(k0_block_data_begin < (karg.K0 - K0PerBlock));
         }
 
         // tail
@@ -519,6 +833,30 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
             blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
         }
+#else
+        // gridwise GEMM pipeline
+        const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
+            (a_b_k0_m_k1_grid_desc.GetLength(I1) * a_b_k0_m_k1_grid_desc.GetLength(I3)) /
+            (K0PerBlock * K1));
+
+        const auto gridwise_gemm_pipeline = GridwiseGemmPipe{};
+
+        gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_b_k0_m_k1_grid_desc,
+                                                               a_b_k0_m_k1_block_desc,
+                                                               a_blockwise_copy,
+                                                               a_grid_buf,
+                                                               a_block_buf,
+                                                               a_block_slice_copy_step,
+                                                               b_b_k0_n_k1_grid_desc,
+                                                               b_b_k0_n_k1_block_desc,
+                                                               b_blockwise_copy,
+                                                               b_grid_buf,
+                                                               b_block_buf,
+                                                               b_block_slice_copy_step,
+                                                               blockwise_gemm,
+                                                               c_thread_buf,
+                                                               num_k_block_main_loop);
+#endif
 
         // output: register to global memory
         {
@@ -647,7 +985,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                 {c_block_desc_mblock_mperblock_nblock_nperblock,
                  make_multi_index(0, 0, 0, 0),
                  c_grid_desc_mblock_mperblock_nblock_nperblock,
-                 make_multi_index(block_work_idx[I1], 0, block_work_idx[I2], 0),
+                 make_multi_index(block_m_id, 0, block_n_id, 0),
                  c_element_op};
 
             constexpr auto mxdlperwave_forward_step =
@@ -716,6 +1054,30 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             });
         }
     }
+
+    static std::string GetTypeString()
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "GemmXdlSplitKCShuffle_"
+            << getGemmSpecializationString(GemmSpec) << "_"
+            << std::string(ALayout::name)[0]
+            << std::string(BLayout::name)[0]
+            << std::string(CLayout::name)[0]
+            << "_"
+            << "B" << BlockSize << "_"
+            << "Vec" << ABlockTransferSrcScalarPerVector << "x"
+            << BBlockTransferSrcScalarPerVector << "x"
+            << CBlockTransferScalarPerVector_NWaveNPerXDL << "_"
+            << MPerBlock << "x"
+            << NPerBlock << "x"
+            << K0PerBlock << "x"
+            << K1 ;
+        // clang-format on
+
+        return str.str();
+    }
 };
 
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
index ffb2926c877e1569847da43fe08c05e30073dc84..d090ba54d88095cb67fa97ad58146a74cef3e346 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -46,7 +46,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
index 7e6dbb3b2ed85f55ec062a7b60cedc7b40337d54..2c09e80fdc7e2987dbb53be557187d05e2a961cf 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -49,7 +49,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
index fb1e34b98543665717624af2201d9cbfe4596cda..c1bc6a8fec37f3ac9d94a371d9426ad108622d3b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -53,7 +53,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_naive_variance.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_normalization_naive_variance.hpp
deleted file mode 100644
index 89efea4d6c36a53fdb98895838848d271857debd..0000000000000000000000000000000000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_naive_variance.hpp
+++ /dev/null
@@ -1,360 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/utility/data_type.hpp"
-#include "ck/utility/reduction_common.hpp"
-#include "ck/utility/reduction_operator.hpp"
-#include "ck/utility/reduction_functions_accumulate.hpp"
-#include "ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp"
-#include "ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp"
-#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-namespace ck {
-
-// Y = Normalization(X, Beta, Gamma)
-template <typename XDataType,
-          typename GammaDataType,
-          typename BetaDataType,
-          typename YDataType,
-          typename AccDataType,
-          typename AccElementwiseOperation,
-          typename GridDesc_M_K,
-          index_t BlockSize,
-          index_t MThreadClusterSize,
-          index_t KThreadClusterSize,
-          index_t MThreadSliceSize,
-          index_t KThreadSliceSize,
-          index_t XSrcVectorDim,
-          index_t XSrcVectorSize,
-          index_t GammaSrcVectorDim,
-          index_t GammaSrcVectorSize,
-          index_t BetaSrcVectorDim,
-          index_t BetaSrcVectorSize,
-          index_t YDstVectorDim,
-          index_t YDstVectorSize,
-          bool SweepOnce>
-struct GridwiseNormalizationNaiveVariance_mk_to_mk
-{
-    static_assert((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
-                      (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
-                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
-
-    static_assert((YDstVectorDim == 0 && MThreadSliceSize % YDstVectorSize == 0) ||
-                      (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
-                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
-
-    static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0);
-
-    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
-
-    using ThreadBufferDimAccessOrder =
-        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
-
-    using ThreadClusterArrangeOrder =
-        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
-
-    static constexpr auto thread_cluster_desc =
-        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
-
-    using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
-        make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{})));
-    using ThreadReduceDstDesc_M =
-        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
-
-    using BlockwiseSumReduce = PartitionedBlockwiseReduction<AccDataType,
-                                                             BlockSize,
-                                                             ThreadClusterLengths_M_K,
-                                                             ThreadClusterArrangeOrder,
-                                                             reduce::Add,
-                                                             true>;
-
-    using ThreadwiseSumReduce = ThreadwiseReduction<AccDataType,
-                                                    ThreadReduceSrcDesc_M_K,
-                                                    ThreadReduceDstDesc_M,
-                                                    reduce::Add,
-                                                    true>;
-
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-
-    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
-    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
-
-    __device__ static void Run(const GridDesc_M_K& x_grid_desc_m_k,
-                               const GridDesc_M_K& gamma_grid_desc_m_k,
-                               const GridDesc_M_K& beta_grid_desc_m_k,
-                               const GridDesc_M_K& y_grid_desc_m_k,
-                               index_t num_k_block_tile_iteration,
-                               AccDataType epsilon,
-                               const XDataType* const __restrict__ p_x_global,
-                               const GammaDataType* const __restrict__ p_gamma_global,
-                               const BetaDataType* const __restrict__ p_beta_global,
-                               YDataType* const __restrict__ p_y_global,
-                               const AccElementwiseOperation acc_elementwise_op)
-    {
-        if constexpr(SweepOnce)
-        {
-            num_k_block_tile_iteration = 1;
-        }
-
-        // LDS
-        __shared__ AccDataType p_reduce_work_buffer[BlockSize];
-
-        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
-
-        auto reduce_work_buf =
-            make_dynamic_buffer<AddressSpaceEnum::Lds>(p_reduce_work_buffer, BlockSize);
-
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize * KThreadSliceSize, true>
-            x_thread_buf;
-
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize * KThreadSliceSize, true>
-            gamma_thread_buf;
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     AccDataType,
-                     MThreadSliceSize * KThreadSliceSize,
-                     true>& beta_thread_buf = gamma_thread_buf;
-
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize * KThreadSliceSize, true>
-            y_thread_buf;
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     AccDataType,
-                     MThreadSliceSize * KThreadSliceSize,
-                     true>& x_square_thread_buf = y_thread_buf;
-
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> mean_thread_buf;
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
-            mean_square_thread_buf;
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>& var_thread_buf =
-            mean_square_thread_buf;
-
-        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
-            mean_thread_buf(I)        = reduce::Add::template GetIdentityValue<AccDataType>();
-            mean_square_thread_buf(I) = reduce::Add::template GetIdentityValue<AccDataType>();
-        });
-
-        const index_t thread_local_id = get_thread_local_1d_id();
-        const index_t block_global_id = get_block_1d_id();
-
-        const auto thread_cluster_idx =
-            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
-
-        const auto thread_m_cluster_id = thread_cluster_idx[I0];
-        const auto thread_k_cluster_id = thread_cluster_idx[I1];
-
-        using ThreadBufferLengths_M_K         = Sequence<MThreadSliceSize, KThreadSliceSize>;
-        constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{}));
-
-        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
-                                                                  AccDataType,
-                                                                  GridDesc_M_K,
-                                                                  decltype(thread_buffer_desc_m_k),
-                                                                  ThreadBufferLengths_M_K,
-                                                                  ThreadBufferDimAccessOrder,
-                                                                  XSrcVectorDim,
-                                                                  XSrcVectorSize,
-                                                                  1,
-                                                                  true>(
-            x_grid_desc_m_k,
-            make_multi_index(block_global_id * M_BlockTileSize +
-                                 thread_m_cluster_id * MThreadSliceSize,
-                             thread_k_cluster_id * KThreadSliceSize));
-
-        auto threadwise_gamma_load =
-            ThreadwiseTensorSliceTransfer_v2<GammaDataType,
-                                             AccDataType,
-                                             GridDesc_M_K,
-                                             decltype(thread_buffer_desc_m_k),
-                                             ThreadBufferLengths_M_K,
-                                             ThreadBufferDimAccessOrder,
-                                             GammaSrcVectorDim,
-                                             GammaSrcVectorSize,
-                                             1,
-                                             true>(
-                gamma_grid_desc_m_k,
-                make_multi_index(block_global_id * M_BlockTileSize +
-                                     thread_m_cluster_id * MThreadSliceSize,
-                                 thread_k_cluster_id * KThreadSliceSize));
-
-        auto threadwise_beta_load =
-            ThreadwiseTensorSliceTransfer_v2<BetaDataType,
-                                             AccDataType,
-                                             GridDesc_M_K,
-                                             decltype(thread_buffer_desc_m_k),
-                                             ThreadBufferLengths_M_K,
-                                             ThreadBufferDimAccessOrder,
-                                             BetaSrcVectorDim,
-                                             BetaSrcVectorSize,
-                                             1,
-                                             true>(
-                beta_grid_desc_m_k,
-                make_multi_index(block_global_id * M_BlockTileSize +
-                                     thread_m_cluster_id * MThreadSliceSize,
-                                 thread_k_cluster_id * KThreadSliceSize));
-
-        auto threadwise_y_store =
-            ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
-                                               YDataType,
-                                               decltype(thread_buffer_desc_m_k),
-                                               GridDesc_M_K,
-                                               AccElementwiseOperation,
-                                               ThreadBufferLengths_M_K,
-                                               ThreadBufferDimAccessOrder,
-                                               YDstVectorDim,
-                                               YDstVectorSize,
-                                               InMemoryDataOperationEnum::Set,
-                                               1,
-                                               true>(
-                y_grid_desc_m_k,
-                make_multi_index(block_global_id * M_BlockTileSize +
-                                     thread_m_cluster_id * MThreadSliceSize,
-                                 thread_k_cluster_id * KThreadSliceSize),
-                acc_elementwise_op);
-
-        // Copy x from Cache
-        // one pass: fwd, second pass: bwd
-        constexpr auto thread_copy_fwd_step_m_k =
-            make_multi_index(0, SweepOnce ? 0 : K_BlockTileSize);
-        constexpr auto thread_copy_bwd_step_m_k =
-            make_multi_index(0, SweepOnce ? 0 : -K_BlockTileSize);
-
-        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_x_global, x_grid_desc_m_k.GetElementSpaceSize());
-
-        const auto gamma_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_gamma_global, gamma_grid_desc_m_k.GetElementSpaceSize());
-
-        const auto beta_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_beta_global, beta_grid_desc_m_k.GetElementSpaceSize());
-
-        // E(x), E[x^2], var(x)
-        // FIXME: Should not hack the transform from deviceOP
-        int reduce_length = x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0];
-
-        index_t reducedTiles = 0;
-        do
-        {
-            threadwise_x_load.Run(x_grid_desc_m_k,
-                                  x_global_val_buf,
-                                  thread_buffer_desc_m_k,
-                                  make_tuple(I0, I0),
-                                  x_thread_buf);
-
-            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                static_for<0, KThreadSliceSize, 1>{}([&](auto iK) {
-                    constexpr auto offset_m_k =
-                        thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK));
-                    x_square_thread_buf(Number<offset_m_k>{}) =
-                        x_thread_buf(Number<offset_m_k>{}) * x_thread_buf(Number<offset_m_k>{});
-                });
-            });
-
-            ThreadwiseSumReduce::Reduce(x_thread_buf, mean_thread_buf);
-            ThreadwiseSumReduce::Reduce(x_square_thread_buf, mean_square_thread_buf);
-
-            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
-
-            ++reducedTiles;
-        } while(reducedTiles < num_k_block_tile_iteration);
-
-        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
-            if constexpr(I > 0)
-                block_sync_lds();
-
-            BlockwiseSumReduce::Reduce(reduce_work_buf, mean_thread_buf(I));
-            mean_thread_buf(I) = mean_thread_buf(I) / reduce_length;
-
-            block_sync_lds();
-
-            BlockwiseSumReduce::Reduce(reduce_work_buf, mean_square_thread_buf(I));
-            mean_square_thread_buf(I) = mean_square_thread_buf(I) / reduce_length;
-
-            // var(x) = E[x^2] - E[x]^2
-            var_thread_buf(I) =
-                mean_square_thread_buf(I) - (mean_thread_buf(I) * mean_thread_buf(I));
-        });
-
-        // y = (x - E[x]) / sqrt(var[x] + epsilon)
-        auto thread_copy_tail_m_k = (num_k_block_tile_iteration - 1) * thread_copy_fwd_step_m_k;
-
-        threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_bwd_step_m_k);
-        threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k, thread_copy_tail_m_k);
-        threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k, thread_copy_tail_m_k);
-        threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_tail_m_k);
-
-        reducedTiles = 0;
-        do
-        {
-            if constexpr(!SweepOnce)
-            {
-                threadwise_x_load.Run(x_grid_desc_m_k,
-                                      x_global_val_buf,
-                                      thread_buffer_desc_m_k,
-                                      make_tuple(I0, I0),
-                                      x_thread_buf);
-            }
-
-            threadwise_gamma_load.Run(gamma_grid_desc_m_k,
-                                      gamma_global_val_buf,
-                                      thread_buffer_desc_m_k,
-                                      make_tuple(I0, I0),
-                                      gamma_thread_buf);
-
-            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                static_for<0, KThreadSliceSize, 1>{}([&](auto iK) {
-                    constexpr auto offset_m_k =
-                        thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK));
-
-                    // normalize
-                    y_thread_buf(Number<offset_m_k>{}) =
-                        (x_thread_buf(Number<offset_m_k>{}) - mean_thread_buf(iM)) /
-                        sqrt(var_thread_buf(iM) + epsilon);
-
-                    // gamma
-                    y_thread_buf(Number<offset_m_k>{}) =
-                        y_thread_buf(Number<offset_m_k>{}) * gamma_thread_buf(Number<offset_m_k>{});
-                });
-            });
-
-            threadwise_beta_load.Run(beta_grid_desc_m_k,
-                                     beta_global_val_buf,
-                                     thread_buffer_desc_m_k,
-                                     make_tuple(I0, I0),
-                                     beta_thread_buf);
-
-            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                static_for<0, KThreadSliceSize, 1>{}([&](auto iK) {
-                    constexpr auto offset_m_k =
-                        thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK));
-
-                    // beta
-                    y_thread_buf(Number<offset_m_k>{}) =
-                        y_thread_buf(Number<offset_m_k>{}) + beta_thread_buf(Number<offset_m_k>{});
-                });
-            });
-
-            threadwise_y_store.Run(thread_buffer_desc_m_k,
-                                   make_tuple(I0, I0),
-                                   y_thread_buf,
-                                   y_grid_desc_m_k,
-                                   y_global_val_buf);
-
-            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_bwd_step_m_k);
-            threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k, thread_copy_bwd_step_m_k);
-            threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k, thread_copy_bwd_step_m_k);
-            threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_bwd_step_m_k);
-
-            ++reducedTiles;
-        } while(reducedTiles < num_k_block_tile_iteration);
-    }
-};
-
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp
index de1ae915920503a086d2cf939ae6e41d27a28838..61d0f9e0d5573c774bbb1dcba707e8c2e775f075 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..8f72f88b068afc4a15e94b16dc659a2b2b9f992a
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+
+namespace ck {
+
+template <typename GridwisePutElementwise1dFunctor,
+          typename InGrid1dDesc,
+          typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation>
+__global__ void kernel_put_element_1d(const InGrid1dDesc in_grid_1d_desc,
+                                      const InDataType* __restrict__ p_in_global,
+                                      const IndexDataType* __restrict__ p_indices_global,
+                                      OutDataType* __restrict__ p_out_global,
+                                      const ElementwiseOperation elementwise_op)
+{
+    GridwisePutElementwise1dFunctor::Run(
+        in_grid_1d_desc, p_in_global, p_indices_global, p_out_global, elementwise_op);
+}
+
+// output[indices] = input
+template <typename InGrid1dDesc,
+          typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation,
+          InMemoryDataOperationEnum MemOp,
+          index_t InVectorSize>
+struct GridwisePutElement_1D
+{
+    static constexpr auto I0 = Number<0>{};
+
+    static constexpr auto thread_buffer_desc_m =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<InVectorSize>{}));
+
+    __device__ static void Run(const InGrid1dDesc& in_grid_1d_desc,
+                               const InDataType* __restrict__ p_in_global,
+                               const IndexDataType* __restrict__ p_indices_global,
+                               OutDataType* __restrict__ p_out_global,
+                               const ElementwiseOperation& elementwise_op)
+    {
+        // Global Memory
+        const auto in_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_in_global, in_grid_1d_desc.GetElementSpaceSize());
+
+        const auto indices_global_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global>(p_indices_global,
+                                                          in_grid_1d_desc.GetElementSpaceSize(),
+                                                          NumericLimits<IndexDataType>::Lowest());
+
+        // VGPR
+        StaticBuffer<AddressSpaceEnum::Vgpr, InDataType, InVectorSize, true> in_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, IndexDataType, InVectorSize, true> indices_thread_buf;
+
+        // Thread id, Block id and index
+        const index_t thread_global_id  = get_thread_global_1d_id();
+        const auto thread_global_offset = make_multi_index(thread_global_id * InVectorSize);
+        const index_t blockSize         = get_block_size();
+        const index_t blockPerGrid      = get_grid_size();
+        const auto M                    = in_grid_1d_desc.GetLength(I0);
+        const index_t loop_step         = blockPerGrid * blockSize * InVectorSize;
+        const auto loop_step_index      = make_multi_index(loop_step);
+
+        auto in_global_load =
+            ThreadwiseTensorSliceTransfer_v2<InDataType,
+                                             InDataType,
+                                             decltype(in_grid_1d_desc),
+                                             decltype(thread_buffer_desc_m),
+                                             Sequence<InVectorSize>, // SliceLengths
+                                             Sequence<0>,            // DimAccessOrder
+                                             0,                      // SrcVectorDim
+                                             InVectorSize,           // ScalarPerVector
+                                             1,                      // SrcScalarStrideInVector
+                                             false>{in_grid_1d_desc, thread_global_offset};
+
+        auto indices_global_load =
+            ThreadwiseTensorSliceTransfer_v2<IndexDataType,
+                                             IndexDataType,
+                                             decltype(in_grid_1d_desc),
+                                             decltype(thread_buffer_desc_m),
+                                             Sequence<InVectorSize>, // SliceLengths
+                                             Sequence<0>,            // DimAccessOrder
+                                             0,                      // SrcVectorDim
+                                             InVectorSize,           // ScalarPerVector
+                                             1,                      // SrcScalarStrideInVector
+                                             false>{in_grid_1d_desc, thread_global_offset};
+
+        index_t num_iter = M / loop_step;
+        do
+        {
+            in_global_load.Run(in_grid_1d_desc,
+                               in_global_buf,
+                               thread_buffer_desc_m,
+                               make_tuple(I0),
+                               in_thread_buf);
+
+            in_global_load.MoveSrcSliceWindow(in_grid_1d_desc, loop_step_index);
+
+            static_for<0, InVectorSize, 1>{}(
+                [&](auto iM) { elementwise_op(in_thread_buf(iM), in_thread_buf[iM]); });
+
+            indices_global_load.Run(in_grid_1d_desc,
+                                    indices_global_buf,
+                                    thread_buffer_desc_m,
+                                    make_tuple(I0),
+                                    indices_thread_buf);
+
+            indices_global_load.MoveSrcSliceWindow(in_grid_1d_desc, loop_step_index);
+
+            static_for<0, InVectorSize, 1>{}([&](auto iM) {
+                if(indices_thread_buf[iM] >= 0)
+                {
+                    if constexpr(MemOp == InMemoryDataOperationEnum::Set)
+                    {
+                        // User should guarantee each index in p_indices_global is different
+                        *(p_out_global + indices_thread_buf[iM]) =
+                            ck::type_convert<OutDataType>(in_thread_buf[iM]);
+                    }
+                    else if constexpr(MemOp == InMemoryDataOperationEnum::AtomicAdd)
+                    {
+                        atomic_add<OutDataType>(p_out_global + indices_thread_buf[iM],
+                                                ck::type_convert<OutDataType>(in_thread_buf[iM]));
+                    }
+                    else if constexpr(MemOp == InMemoryDataOperationEnum::AtomicMax)
+                    {
+                        atomic_max<OutDataType>(p_out_global + indices_thread_buf[iM],
+                                                ck::type_convert<OutDataType>(in_thread_buf[iM]));
+                    }
+                    else if constexpr(MemOp == InMemoryDataOperationEnum::Add)
+                    {
+                        // User should guarantee each index in p_indices_global is different
+                        *(p_out_global + indices_thread_buf[iM]) +=
+                            ck::type_convert<OutDataType>(in_thread_buf[iM]);
+                    }
+                    else
+                    {
+                        static_assert(MemOp == InMemoryDataOperationEnum::Set ||
+                                      MemOp == InMemoryDataOperationEnum::AtomicAdd ||
+                                      MemOp == InMemoryDataOperationEnum::AtomicMax ||
+                                      MemOp == InMemoryDataOperationEnum::Add);
+                    }
+                }
+            });
+
+        } while(--num_iter);
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp
index 901e7aee98a7b6ab1c37b7dd4dc36f796632fef6..41352fabeb361b017949738bda31649a3ce0dd40 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp
index 88c7b6acfeb9c43ec0e8b5a5ab557f834430fb40..0ad36b418a42bff56eb87ba35bb5150808e9cac4 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp
index 0344e68305b1a951cfc41a53666090bd1e0113b6..5f56ac6fc4664be1685a950dc6e242b3e256e178 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embedding3_forward_layernorm.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
similarity index 69%
rename from include/ck/tensor_operation/gpu/grid/gridwise_sparse_embedding3_forward_layernorm.hpp
rename to include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
index 3de6aa08c45b9fadbdf88d1554d40824a5f555ce..ee68660a06a2933e45ead9c05fadc674ec9d5ae7 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embedding3_forward_layernorm.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,33 +17,24 @@ template <typename GridwiseSparseEmbedding,
           typename BetaDataType,
           typename AccDataType,
           typename OutType,
-          typename OutGridDesc>
+          typename OutGridDesc,
+          typename EmbElementwiseOperation,
+          ck::index_t NumEmbeddings>
 #if CK_USE_LAUNCH_BOUNDS
 __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-    __global__ void kernel_sparse_embedding3_forward_layernorm(OutType* p_out,
-                                                               const EmbType* p_emb_a,
-                                                               const EmbType* p_emb_b,
-                                                               const EmbType* p_emb_c,
-                                                               const IndexType* p_index_a,
-                                                               const IndexType* p_index_b,
-                                                               const IndexType* p_index_c,
-                                                               const GammaDataType* p_gamma,
-                                                               const BetaDataType* p_beta,
-                                                               const OutGridDesc out_grid_desc,
-                                                               const AccDataType epsilon)
+    __global__ void kernel_sparse_embeddings_forward_layernorm(
+        OutType* p_out,
+        const ck::Array<EmbType*, NumEmbeddings> p_embs,
+        const ck::Array<IndexType*, NumEmbeddings> p_indexes,
+        const GammaDataType* p_gamma,
+        const BetaDataType* p_beta,
+        const OutGridDesc out_grid_desc,
+        const AccDataType epsilon,
+        const EmbElementwiseOperation emb_elementwise_op)
 {
-    GridwiseSparseEmbedding::Run(p_out,
-                                 p_emb_a,
-                                 p_emb_b,
-                                 p_emb_c,
-                                 p_index_a,
-                                 p_index_b,
-                                 p_index_c,
-                                 p_gamma,
-                                 p_beta,
-                                 out_grid_desc,
-                                 epsilon);
+    GridwiseSparseEmbedding::Run(
+        p_out, p_embs, p_indexes, p_gamma, p_beta, out_grid_desc, epsilon, emb_elementwise_op);
 }
 
 template <typename EmbType,
@@ -53,14 +44,16 @@ template <typename EmbType,
           typename AccDataType,
           typename OutType,
           typename OutGridDesc,
+          typename EmbElementwiseOperation,
           ck::index_t BlockSize,
           ck::index_t DimClusterSize,
           ck::index_t RowClusterSize,
           ck::index_t DimPerBlock,   // Row x Dim, along Dim
           ck::index_t RowPerBlock,   // Row x Dim, along Row
           ck::index_t DimThreadSize, // this is actually not vector, but number of registers
-          ck::index_t RowVectorSize>
-struct GridwiseSparseEmbedding3ForwardLayernorm
+          ck::index_t RowVectorSize,
+          ck::index_t NumEmbeddings>
+struct GridwiseSparseEmbeddingsForwardLayernorm
 {
     static constexpr auto I0          = Number<0>{};
     static constexpr auto I1          = Number<1>{};
@@ -97,23 +90,17 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
         BlockwiseWelford<AccDataType, BlockSize, ThreadClusterLength, Sequence<0, 1>>;
 
     __device__ static void Run(OutType* p_out,
-                               const EmbType* p_emb_a,
-                               const EmbType* p_emb_b,
-                               const EmbType* p_emb_c,
-                               const IndexType* p_index_a,
-                               const IndexType* p_index_b,
-                               const IndexType* p_index_c,
+                               const ck::Array<EmbType*, NumEmbeddings> p_embs,
+                               const ck::Array<IndexType*, NumEmbeddings> p_indexes,
                                const GammaDataType* p_gamma,
                                const BetaDataType* p_beta,
                                const OutGridDesc,
-                               const AccDataType epsilon)
+                               const AccDataType epsilon,
+                               const EmbElementwiseOperation emb_elementwise_op)
     {
         const index_t thread_local_id = get_thread_local_1d_id();
         const index_t block_global_id = get_block_1d_id();
 
-        // const auto index_length = out_grid_desc.GetLength(I0);
-        // const auto emb_dim      = out_grid_desc.GetLength(I1);
-
         constexpr auto thread_cluster_desc =
             make_cluster_descriptor(Sequence<DimClusterSize, RowClusterSize>{}, Sequence<0, 1>{});
 
@@ -141,13 +128,11 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
         constexpr auto gamma_beta_buf_desc =
             make_naive_tensor_descriptor_packed(make_tuple(RowSubBlocks, RowVectorSize));
 
-        StaticBuffer<AddressSpaceEnum::Vgpr, EmbType, thread_buf_size, true> in_thread_buf_a;
-        StaticBuffer<AddressSpaceEnum::Vgpr, EmbType, thread_buf_size, true> in_thread_buf_b;
-        StaticBuffer<AddressSpaceEnum::Vgpr, EmbType, thread_buf_size, true> in_thread_buf_c;
-
-        StaticBuffer<AddressSpaceEnum::Sgpr, IndexType, DimPerBlock, true> index_buf_a;
-        StaticBuffer<AddressSpaceEnum::Sgpr, IndexType, DimPerBlock, true> index_buf_b;
-        StaticBuffer<AddressSpaceEnum::Sgpr, IndexType, DimPerBlock, true> index_buf_c;
+        ck::Array<StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, thread_buf_size, true>,
+                  NumEmbeddings>
+            in_thread_bufs;
+        ck::Array<StaticBuffer<AddressSpaceEnum::Vgpr, IndexType, DimPerBlock, true>, NumEmbeddings>
+            index_bufs;
 
         StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, thread_buf_size, true> acc_thread_buf;
 
@@ -160,42 +145,31 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
         StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, mean_var_buf_size, true> var_thread_buf;
 
         auto load_current_sub_row = [&](auto i_dim_sub_, auto i_row_sub_) {
-            vector_type_maker_t<EmbType, RowVectorSize> emb_vector_a;
-            vector_type_maker_t<EmbType, RowVectorSize> emb_vector_b;
-            vector_type_maker_t<EmbType, RowVectorSize> emb_vector_c;
-
-            using src_vector_t = typename decltype(emb_vector_a)::type;
+            ck::Array<vector_type_maker_t<EmbType, RowVectorSize>, NumEmbeddings> emb_vectors;
+            auto emb_a         = emb_vectors[0];
+            using src_vector_t = typename decltype(emb_a)::type;
             static_for<0, DimThreadSize, 1>{}([&](auto i_dim_vec_) {
                 constexpr auto current_dim = i_dim_sub_ * DimPerSubBlock + i_dim_vec_;
-                IndexType index_a          = index_buf_a[Number<current_dim>{}];
-                IndexType index_b          = index_buf_b[Number<current_dim>{}];
-                IndexType index_c          = index_buf_c[Number<current_dim>{}];
 
                 auto thread_offset = (thread_row_cluster_id + i_row_sub_ * RowClusterSize) *
                                      sizeof(EmbType) * RowVectorSize;
+                static_for<0, NumEmbeddings, 1>{}([&](auto i_embedding_) {
+                    IndexType index = index_bufs[i_embedding_][Number<current_dim>{}];
 
-                int32x4_t emb_res_a =
-                    make_wave_buffer_resource_with_default_range(p_emb_a + index_a * RowPerBlock);
-                int32x4_t emb_res_b =
-                    make_wave_buffer_resource_with_default_range(p_emb_b + index_b * RowPerBlock);
-                int32x4_t emb_res_c =
-                    make_wave_buffer_resource_with_default_range(p_emb_c + index_c * RowPerBlock);
-                emb_vector_a.template AsType<src_vector_t>()(I0) =
-                    amd_buffer_load_impl<EmbType, RowVectorSize>(emb_res_a, thread_offset, 0);
-                emb_vector_b.template AsType<src_vector_t>()(I0) =
-                    amd_buffer_load_impl<EmbType, RowVectorSize>(emb_res_b, thread_offset, 0);
-                emb_vector_c.template AsType<src_vector_t>()(I0) =
-                    amd_buffer_load_impl<EmbType, RowVectorSize>(emb_res_c, thread_offset, 0);
+                    int32x4_t emb_res = make_wave_buffer_resource_with_default_range(
+                        p_embs[i_embedding_] + index * RowPerBlock);
+                    emb_vectors(i_embedding_).template AsType<src_vector_t>()(I0) =
+                        amd_buffer_load_impl<EmbType, RowVectorSize>(emb_res, thread_offset, 0);
+                });
 
                 static_for<0, RowVectorSize, 1>{}([&](auto i_row_vec_) {
                     constexpr auto register_offset = thread_buf_desc.CalculateOffset(
                         make_tuple(i_dim_sub_, i_dim_vec_, i_row_sub_, i_row_vec_));
-                    in_thread_buf_a(Number<register_offset>{}) =
-                        emb_vector_a.template AsType<EmbType>()[i_row_vec_];
-                    in_thread_buf_b(Number<register_offset>{}) =
-                        emb_vector_b.template AsType<EmbType>()[i_row_vec_];
-                    in_thread_buf_c(Number<register_offset>{}) =
-                        emb_vector_c.template AsType<EmbType>()[i_row_vec_];
+                    static_for<0, NumEmbeddings, 1>{}([&](auto i_embedding_) {
+                        in_thread_bufs(i_embedding_)(Number<register_offset>{}) =
+                            ck::type_convert<AccDataType>(
+                                emb_vectors[i_embedding_].template AsType<EmbType>()[i_row_vec_]);
+                    });
                 });
             });
         };
@@ -205,14 +179,15 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
                 static_for<0, RowVectorSize, 1>{}([&](auto i_row_vec_) {
                     constexpr auto register_offset = thread_buf_desc.CalculateOffset(
                         make_tuple(i_dim_sub_, i_dim_vec_, i_row_sub_, i_row_vec_));
-                    AccDataType va =
-                        ck::type_convert<AccDataType>(in_thread_buf_a(Number<register_offset>{}));
-                    AccDataType vb =
-                        ck::type_convert<AccDataType>(in_thread_buf_b(Number<register_offset>{}));
-                    AccDataType vc =
-                        ck::type_convert<AccDataType>(in_thread_buf_c(Number<register_offset>{}));
-
-                    acc_thread_buf(Number<register_offset>{}) += va + vb + vc;
+                    auto in_data_refs = generate_tie(
+                        [&](auto i_embedding_) -> const auto& {
+                            return in_thread_bufs(i_embedding_)(Number<register_offset>{});
+                        },
+                        Number<NumEmbeddings>{});
+                    auto out_data_refs = generate_tie(
+                        [&](auto) -> auto& { return acc_thread_buf(Number<register_offset>{}); },
+                        Number<1>{});
+                    unpack2(emb_elementwise_op, out_data_refs, in_data_refs);
                 });
             });
         };
@@ -242,7 +217,8 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
 
                 constexpr auto mean_var_offset =
                     mean_var_buf_desc.CalculateOffset(make_tuple(i_dim_sub_, i_dim_vec_));
-
+                auto divisor =
+                    1 / __builtin_amdgcn_sqrtf(var_thread_buf(Number<mean_var_offset>{}) + epsilon);
                 static_for<0, RowVectorSize, 1>{}([&](auto i_row_vec_) {
                     constexpr auto register_offset = thread_buf_desc.CalculateOffset(
                         make_tuple(i_dim_sub_, i_dim_vec_, i_row_sub_, i_row_vec_));
@@ -250,9 +226,8 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
                         gamma_beta_buf_desc.CalculateOffset(make_tuple(i_row_sub_, i_row_vec_));
 
                     auto acc_val = acc_thread_buf[Number<register_offset>{}];
-                    acc_val      = (acc_val - mean_thread_buf(Number<mean_var_offset>{})) /
-                              sqrt(var_thread_buf(Number<mean_var_offset>{}) + epsilon);
-                    acc_val = acc_val * gamma_thread_buf[Number<gamma_beta_offset>{}] +
+                    acc_val      = (acc_val - mean_thread_buf(Number<mean_var_offset>{})) * divisor;
+                    acc_val      = acc_val * gamma_thread_buf[Number<gamma_beta_offset>{}] +
                               beta_thread_buf[Number<gamma_beta_offset>{}];
 
                     out_vector.template AsType<OutType>()(Number<i_row_vec_>{}) =
@@ -273,9 +248,10 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
         // first load index
         ck::static_for<0, DimPerBlock, 1>{}([&](auto i_idx_) {
             // prefer use s_load
-            index_buf_a(i_idx_) = p_index_a[index_start + i_idx_.value];
-            index_buf_b(i_idx_) = p_index_b[index_start + i_idx_.value];
-            index_buf_c(i_idx_) = p_index_c[index_start + i_idx_.value];
+            ck::static_for<0, NumEmbeddings, 1>{}([&](auto i_embedding_) {
+                index_bufs(i_embedding_)(i_idx_) =
+                    p_indexes[i_embedding_][index_start + i_idx_.value];
+            });
         });
 
         // load gamma/beta
@@ -329,7 +305,6 @@ struct GridwiseSparseEmbedding3ForwardLayernorm
             static_for<0, mean_var_buf_size, 1>{}([&](auto I) {
                 if constexpr(I > 0)
                     block_sync_lds();
-
                 BlockwiseWelford::Run(
                     mean_thread_buf(I), var_thread_buf(I), threadwise_welford.cur_count_);
             });
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..c3f122106df29a471e376c453d055806c9aa2514
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
@@ -0,0 +1,505 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+
+#include "ck/utility/reduction_operator.hpp"
+#include "ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp"
+#include "ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+// Y = Normalization(X, Beta, Gamma)
+template <typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
+          typename GridDesc_M_K,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcVectorDim,
+          index_t XSrcVectorSize,
+          index_t GammaSrcVectorDim,
+          index_t GammaSrcVectorSize,
+          index_t BetaSrcVectorDim,
+          index_t BetaSrcVectorSize,
+          index_t YDstVectorDim,
+          index_t YDstVectorSize,
+          bool SweepOnce>
+struct GridwiseNormalizationNaiveVariance_mk_to_mk
+{
+    static_assert((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert((YDstVectorDim == 0 && MThreadSliceSize % YDstVectorSize == 0) ||
+                      (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert(XSrcVectorSize == YDstVectorSize);
+    static_assert(XSrcVectorSize == GammaSrcVectorSize);
+    static_assert(XSrcVectorSize == BetaSrcVectorSize);
+
+    static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0);
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using ThreadBufferDimAccessOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadBufferLengths_M_K                = Sequence<MThreadSliceSize, XSrcVectorSize>;
+    static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
+
+    using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{})));
+    using ThreadReduceDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using BlockwiseSumReduce = PartitionedBlockwiseReduction<ComputeDataType,
+                                                             BlockSize,
+                                                             ThreadClusterLengths_M_K,
+                                                             ThreadClusterArrangeOrder,
+                                                             reduce::Add,
+                                                             true>;
+
+    using ThreadwiseSumReduce = ThreadwiseReduction<ComputeDataType,
+                                                    ThreadReduceSrcDesc_M_K,
+                                                    ThreadReduceDstDesc_M,
+                                                    reduce::Add,
+                                                    true>;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+
+    static constexpr index_t M_BlockTileSize     = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize     = KThreadClusterSize * KThreadSliceSize;
+    static constexpr index_t K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize;
+
+    static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
+
+    __device__ static void Run(const GridDesc_M_K& x_grid_desc_m_k,
+                               const GridDesc_M_K& gamma_grid_desc_m_k,
+                               const GridDesc_M_K& beta_grid_desc_m_k,
+                               const GridDesc_M_K& y_grid_desc_m_k,
+                               index_t num_k_block_tile_iteration,
+                               ComputeDataType epsilon,
+                               const XDataType* const __restrict__ p_x_global,
+                               const GammaDataType* const __restrict__ p_gamma_global,
+                               const BetaDataType* const __restrict__ p_beta_global,
+                               YDataType* const __restrict__ p_y_global,
+                               const YElementwiseOperation y_elementwise_op)
+    {
+        // LDS
+        __shared__ ComputeDataType p_reduce_work_buffer[BlockSize];
+
+        auto reduce_work_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Lds>(p_reduce_work_buffer, BlockSize);
+
+        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
+
+        auto x_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * XSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        auto gamma_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * GammaSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        auto& beta_thread_buf = gamma_thread_buf;
+
+        auto y_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * YDstVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        auto& x_square_thread_buf = y_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            mean_square_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>&
+            var_thread_buf = mean_square_thread_buf;
+
+        const index_t thread_local_id = get_thread_local_1d_id();
+        const index_t block_global_id = get_block_1d_id();
+
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  ComputeDataType,
+                                                                  GridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  ThreadBufferDimAccessOrder,
+                                                                  XSrcVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize,
+                             thread_k_cluster_id * XSrcVectorSize));
+
+        auto threadwise_gamma_load =
+            ThreadwiseTensorSliceTransfer_v2<GammaDataType,
+                                             ComputeDataType,
+                                             GridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             ThreadBufferDimAccessOrder,
+                                             GammaSrcVectorDim,
+                                             GammaSrcVectorSize,
+                                             1,
+                                             true>(
+                gamma_grid_desc_m_k,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * GammaSrcVectorSize));
+
+        auto threadwise_beta_load =
+            ThreadwiseTensorSliceTransfer_v2<BetaDataType,
+                                             ComputeDataType,
+                                             GridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             ThreadBufferDimAccessOrder,
+                                             BetaSrcVectorDim,
+                                             BetaSrcVectorSize,
+                                             1,
+                                             true>(
+                beta_grid_desc_m_k,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * BetaSrcVectorSize));
+
+        auto threadwise_y_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               YDataType,
+                                               decltype(thread_buffer_desc_m_k),
+                                               GridDesc_M_K,
+                                               YElementwiseOperation,
+                                               ThreadBufferLengths_M_K,
+                                               ThreadBufferDimAccessOrder,
+                                               YDstVectorDim,
+                                               YDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                y_grid_desc_m_k,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * YDstVectorSize),
+                y_elementwise_op);
+
+        constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
+        constexpr auto thread_copy_bwd_step_m_k =
+            make_multi_index(0, SweepOnce ? 0 : -K_BlockTileSize);
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x_global, x_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto gamma_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_gamma_global, gamma_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto beta_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_beta_global, beta_grid_desc_m_k.GetElementSpaceSize());
+
+        // E(x), E[x^2], var(x)
+        // FIXME: Should not hack the transform from deviceOP
+        int reduce_length = x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0];
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I)        = reduce::Add::template GetIdentityValue<ComputeDataType>();
+            mean_square_thread_buf(I) = reduce::Add::template GetIdentityValue<ComputeDataType>();
+        });
+
+        // Separate sweep once and sweep twice pipeline
+        if constexpr(SweepOnce)
+        {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_x_load.Run(x_grid_desc_m_k,
+                                      x_global_val_buf,
+                                      thread_buffer_desc_m_k,
+                                      make_tuple(I0, I0),
+                                      x_thread_buf(i));
+
+                threadwise_gamma_load.Run(gamma_grid_desc_m_k,
+                                          gamma_global_val_buf,
+                                          thread_buffer_desc_m_k,
+                                          make_tuple(I0, I0),
+                                          gamma_thread_buf(i));
+
+                static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                    static_for<0, XSrcVectorSize, 1>{}([&](auto iK) {
+                        constexpr auto offset_m_k =
+                            thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK));
+                        x_square_thread_buf(i)(Number<offset_m_k>{}) =
+                            x_thread_buf(i)(Number<offset_m_k>{}) *
+                            x_thread_buf(i)(Number<offset_m_k>{});
+                    });
+                });
+
+                ThreadwiseSumReduce::Reduce(x_thread_buf[i], mean_thread_buf);
+                ThreadwiseSumReduce::Reduce(x_square_thread_buf[i], mean_square_thread_buf);
+
+                if constexpr(i != ThreadBufferNumber - 1)
+                {
+                    threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+                    threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                             thread_copy_fwd_step_m_k);
+                }
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                if constexpr(I > 0)
+                    block_sync_lds();
+
+                BlockwiseSumReduce::Reduce(reduce_work_buf, mean_thread_buf(I));
+                mean_thread_buf(I) = mean_thread_buf(I) / reduce_length;
+
+                block_sync_lds();
+
+                BlockwiseSumReduce::Reduce(reduce_work_buf, mean_square_thread_buf(I));
+                mean_square_thread_buf(I) = mean_square_thread_buf(I) / reduce_length;
+
+                // var(x) = E[x^2] - E[x]^2
+                var_thread_buf(I) =
+                    mean_square_thread_buf(I) - (mean_thread_buf(I) * mean_thread_buf(I));
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                    static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                        constexpr auto offset_m_k =
+                            thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                        // normalize
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
+                            divisor;
+
+                        // gamma & beta
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) *
+                            gamma_thread_buf(iK0)(Number<offset_m_k>{});
+                    });
+                });
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_beta_load.Run(beta_grid_desc_m_k,
+                                         beta_global_val_buf,
+                                         thread_buffer_desc_m_k,
+                                         make_tuple(I0, I0),
+                                         beta_thread_buf(i));
+
+                if constexpr(i != ThreadBufferNumber - 1)
+                    threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                            thread_copy_fwd_step_m_k);
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                    static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                        constexpr auto offset_m_k =
+                            thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                        // beta
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) +
+                            beta_thread_buf(iK0)(Number<offset_m_k>{});
+                    });
+                });
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_y_store.Run(thread_buffer_desc_m_k,
+                                       make_tuple(I0, I0),
+                                       y_thread_buf(i),
+                                       y_grid_desc_m_k,
+                                       y_global_val_buf);
+
+                if constexpr(i != ThreadBufferNumber - 1)
+                    threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k,
+                                                          thread_copy_fwd_step_m_k);
+            });
+        } // end of sweep once
+        else
+        {
+            for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+            {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_x_load.Run(x_grid_desc_m_k,
+                                          x_global_val_buf,
+                                          thread_buffer_desc_m_k,
+                                          make_tuple(I0, I0),
+                                          x_thread_buf(i));
+                    threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+
+                    static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                        static_for<0, XSrcVectorSize, 1>{}([&](auto iK) {
+                            constexpr auto offset_m_k =
+                                thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK));
+                            x_square_thread_buf(i)(Number<offset_m_k>{}) =
+                                x_thread_buf(i)(Number<offset_m_k>{}) *
+                                x_thread_buf(i)(Number<offset_m_k>{});
+                        });
+                    });
+
+                    ThreadwiseSumReduce::Reduce(x_thread_buf[i], mean_thread_buf);
+                    ThreadwiseSumReduce::Reduce(x_square_thread_buf[i], mean_square_thread_buf);
+                });
+            }
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                if constexpr(I > 0)
+                    block_sync_lds();
+
+                BlockwiseSumReduce::Reduce(reduce_work_buf, mean_thread_buf(I));
+                mean_thread_buf(I) = mean_thread_buf(I) / reduce_length;
+
+                block_sync_lds();
+
+                BlockwiseSumReduce::Reduce(reduce_work_buf, mean_square_thread_buf(I));
+                mean_square_thread_buf(I) = mean_square_thread_buf(I) / reduce_length;
+
+                // var(x) = E[x^2] - E[x]^2
+                var_thread_buf(I) =
+                    mean_square_thread_buf(I) - (mean_thread_buf(I) * mean_thread_buf(I));
+            });
+
+            auto thread_copy_tail_m_k =
+                (num_k_block_tile_iteration - 1) * ThreadBufferNumber * thread_copy_fwd_step_m_k;
+
+            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_bwd_step_m_k);
+            threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k, thread_copy_tail_m_k);
+            threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k, thread_copy_tail_m_k);
+            threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_tail_m_k);
+
+            for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+            {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_x_load.Run(x_grid_desc_m_k,
+                                          x_global_val_buf,
+                                          thread_buffer_desc_m_k,
+                                          make_tuple(I0, I0),
+                                          x_thread_buf(i));
+                    threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+                });
+
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_gamma_load.Run(gamma_grid_desc_m_k,
+                                              gamma_global_val_buf,
+                                              thread_buffer_desc_m_k,
+                                              make_tuple(I0, I0),
+                                              gamma_thread_buf(i));
+
+                    threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                             thread_copy_fwd_step_m_k);
+                });
+
+                static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                    auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
+                    static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                        static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                            constexpr auto offset_m_k =
+                                thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                            // normalize
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                                (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
+                                divisor;
+
+                            // gamma
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                                y_thread_buf(iK0)(Number<offset_m_k>{}) *
+                                gamma_thread_buf(iK0)(Number<offset_m_k>{});
+                        });
+                    });
+                });
+
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_beta_load.Run(beta_grid_desc_m_k,
+                                             beta_global_val_buf,
+                                             thread_buffer_desc_m_k,
+                                             make_tuple(I0, I0),
+                                             beta_thread_buf(i));
+                    threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                            thread_copy_fwd_step_m_k);
+                });
+
+                static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                    static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                        static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                            constexpr auto offset_m_k =
+                                thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                            // beta
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                                y_thread_buf(iK0)(Number<offset_m_k>{}) +
+                                beta_thread_buf(iK0)(Number<offset_m_k>{});
+                        });
+                    });
+                });
+
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_y_store.Run(thread_buffer_desc_m_k,
+                                           make_tuple(I0, I0),
+                                           y_thread_buf(i),
+                                           y_grid_desc_m_k,
+                                           y_global_val_buf);
+                    threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k,
+                                                          thread_copy_fwd_step_m_k);
+                });
+
+                threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, 2 * thread_copy_bwd_step_m_k);
+                threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                         2 * thread_copy_bwd_step_m_k);
+                threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                        2 * thread_copy_bwd_step_m_k);
+                threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k,
+                                                      2 * thread_copy_bwd_step_m_k);
+            }
+        } // end of sweep twice
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..e50fb9813325b08576f49d321e7a5de5d51bbff2
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
@@ -0,0 +1,195 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp"
+
+namespace ck {
+template <typename GridwiseReduction,
+          typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
+          typename GridDesc_M_K>
+__global__ void kernel_normalization(const GridDesc_M_K x_grid_desc_m_k,
+                                     const GridDesc_M_K gamma_grid_desc_m_k,
+                                     const GridDesc_M_K beta_grid_desc_m_k,
+                                     const GridDesc_M_K y_grid_desc_m_k,
+                                     index_t num_k_block_tile_iteration,
+                                     ComputeDataType epsilon,
+                                     const XDataType* const __restrict__ p_x_global,
+                                     const GammaDataType* const __restrict__ p_gamma_global,
+                                     const BetaDataType* const __restrict__ p_beta_global,
+                                     YDataType* const __restrict__ p_y_global,
+                                     const YElementwiseOperation y_elementwise_op)
+{
+    GridwiseReduction::Run(x_grid_desc_m_k,
+                           gamma_grid_desc_m_k,
+                           beta_grid_desc_m_k,
+                           y_grid_desc_m_k,
+                           num_k_block_tile_iteration,
+                           epsilon,
+                           p_x_global,
+                           p_gamma_global,
+                           p_beta_global,
+                           p_y_global,
+                           y_elementwise_op);
+};
+
+template <typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
+          typename GridDesc_M_K,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcVectorDim,
+          index_t XSrcVectorSize,
+          index_t GammaSrcVectorDim,
+          index_t GammaSrcVectorSize,
+          index_t BetaSrcVectorDim,
+          index_t BetaSrcVectorSize,
+          index_t YDstVectorDim,
+          index_t YDstVectorSize,
+          bool UseWelford>
+auto NormalizationKernelSelector(bool isSweepOnce)
+{
+    using GridwiseNormalizationGenericNaive =
+        GridwiseNormalizationNaiveVariance_mk_to_mk<XDataType,
+                                                    GammaDataType,
+                                                    BetaDataType,
+                                                    YDataType,
+                                                    ComputeDataType,
+                                                    YElementwiseOperation,
+                                                    GridDesc_M_K,
+                                                    BlockSize,
+                                                    MThreadClusterSize,
+                                                    KThreadClusterSize,
+                                                    MThreadSliceSize,
+                                                    KThreadSliceSize,
+                                                    XSrcVectorDim,
+                                                    XSrcVectorSize,
+                                                    GammaSrcVectorDim,
+                                                    GammaSrcVectorSize,
+                                                    BetaSrcVectorDim,
+                                                    BetaSrcVectorSize,
+                                                    YDstVectorDim,
+                                                    YDstVectorSize,
+                                                    false>;
+    using GridwiseNormalizationSweepOnceNaive =
+        GridwiseNormalizationNaiveVariance_mk_to_mk<XDataType,
+                                                    GammaDataType,
+                                                    BetaDataType,
+                                                    YDataType,
+                                                    ComputeDataType,
+                                                    YElementwiseOperation,
+                                                    GridDesc_M_K,
+                                                    BlockSize,
+                                                    MThreadClusterSize,
+                                                    KThreadClusterSize,
+                                                    MThreadSliceSize,
+                                                    KThreadSliceSize,
+                                                    XSrcVectorDim,
+                                                    XSrcVectorSize,
+                                                    GammaSrcVectorDim,
+                                                    GammaSrcVectorSize,
+                                                    BetaSrcVectorDim,
+                                                    BetaSrcVectorSize,
+                                                    YDstVectorDim,
+                                                    YDstVectorSize,
+                                                    true>;
+    using GridwiseNormalizationGenericWelford =
+        GridwiseNormalizationWelfordVariance_mk_to_mk<XDataType,
+                                                      GammaDataType,
+                                                      BetaDataType,
+                                                      YDataType,
+                                                      ComputeDataType,
+                                                      YElementwiseOperation,
+                                                      GridDesc_M_K,
+                                                      BlockSize,
+                                                      MThreadClusterSize,
+                                                      KThreadClusterSize,
+                                                      MThreadSliceSize,
+                                                      KThreadSliceSize,
+                                                      XSrcVectorDim,
+                                                      XSrcVectorSize,
+                                                      GammaSrcVectorDim,
+                                                      GammaSrcVectorSize,
+                                                      BetaSrcVectorDim,
+                                                      BetaSrcVectorSize,
+                                                      YDstVectorDim,
+                                                      YDstVectorSize,
+                                                      false>;
+    using GridwiseNormalizationSweepOnceWelford =
+        GridwiseNormalizationWelfordVariance_mk_to_mk<XDataType,
+                                                      GammaDataType,
+                                                      BetaDataType,
+                                                      YDataType,
+                                                      ComputeDataType,
+                                                      YElementwiseOperation,
+                                                      GridDesc_M_K,
+                                                      BlockSize,
+                                                      MThreadClusterSize,
+                                                      KThreadClusterSize,
+                                                      MThreadSliceSize,
+                                                      KThreadSliceSize,
+                                                      XSrcVectorDim,
+                                                      XSrcVectorSize,
+                                                      GammaSrcVectorDim,
+                                                      GammaSrcVectorSize,
+                                                      BetaSrcVectorDim,
+                                                      BetaSrcVectorSize,
+                                                      YDstVectorDim,
+                                                      YDstVectorSize,
+                                                      true>;
+
+    if constexpr(UseWelford)
+    {
+        return isSweepOnce ? kernel_normalization<GridwiseNormalizationSweepOnceWelford,
+                                                  XDataType,
+                                                  GammaDataType,
+                                                  BetaDataType,
+                                                  YDataType,
+                                                  ComputeDataType,
+                                                  YElementwiseOperation,
+                                                  GridDesc_M_K>
+                           : kernel_normalization<GridwiseNormalizationGenericWelford,
+                                                  XDataType,
+                                                  GammaDataType,
+                                                  BetaDataType,
+                                                  YDataType,
+                                                  ComputeDataType,
+                                                  YElementwiseOperation,
+                                                  GridDesc_M_K>;
+    }
+    else
+    {
+        return isSweepOnce ? kernel_normalization<GridwiseNormalizationSweepOnceNaive,
+                                                  XDataType,
+                                                  GammaDataType,
+                                                  BetaDataType,
+                                                  YDataType,
+                                                  ComputeDataType,
+                                                  YElementwiseOperation,
+                                                  GridDesc_M_K>
+                           : kernel_normalization<GridwiseNormalizationGenericNaive,
+                                                  XDataType,
+                                                  GammaDataType,
+                                                  BetaDataType,
+                                                  YDataType,
+                                                  ComputeDataType,
+                                                  YElementwiseOperation,
+                                                  GridDesc_M_K>;
+    }
+}
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..fc42e9762997d0c554130beb3d73ac0eebf16476
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
@@ -0,0 +1,252 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/math.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename XDataType,
+          typename ComputeDataType,
+          typename MeanVarDataType,
+          typename XGridDesc_M_K,
+          typename MeanVarGridDesc_M_KBlock,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcVectorDim,
+          index_t XSrcVectorSize>
+struct GridwiseNormalizationSplitK1st
+{
+    static_assert((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0);
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using ThreadBufferDimAccessOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadBufferLengths_M_K                = Sequence<MThreadSliceSize, XSrcVectorSize>;
+    static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
+
+    using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
+    static constexpr auto thread_buffer_desc_m_1 =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, I1));
+
+    using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{})));
+    using ThreadReduceDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using ThreadwiseWelford =
+        ThreadwiseWelford<ComputeDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
+
+    using BlockwiseWelford = BlockwiseWelford<ComputeDataType,
+                                              BlockSize,
+                                              ThreadClusterLengths_M_K,
+                                              ThreadClusterArrangeOrder,
+                                              false>;
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    static constexpr index_t M_BlockTileSize     = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize     = KThreadClusterSize * KThreadSliceSize;
+    static constexpr index_t K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize;
+
+    static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
+
+    __device__ static int
+    GetKPerThread(int kRaw, int kGridSize, int block_k_cluster_id, int thread_k_cluster_id)
+    {
+        bool is_rightmost_block = block_k_cluster_id == kGridSize - 1;
+
+        if(is_rightmost_block)
+        {
+            int left_kPerBlock = math::integer_divide_ceil(kRaw, kGridSize);
+            int kPerBlock      = kRaw % kGridSize == 0 ? left_kPerBlock : kRaw % left_kPerBlock;
+            int kPerThread =
+                kPerBlock < K_BlockTileSize ? 0 : KThreadSliceSize * (kPerBlock / K_BlockTileSize);
+            int kPerBlockTail = kPerBlock - kPerThread * KThreadClusterSize;
+
+            if(kPerBlockTail > 0)
+            {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    int thread_max_len =
+                        (thread_k_cluster_id + 1) * XSrcVectorSize + K_BlockTileStepSize * i;
+                    int delta = thread_max_len - kPerBlockTail;
+                    delta     = math::clamp(thread_max_len - kPerBlockTail, 0, XSrcVectorSize);
+                    kPerThread += XSrcVectorSize - delta;
+                });
+            }
+
+            return kPerThread;
+        }
+        else
+        {
+            int kPerBlock = math::integer_divide_ceil(kRaw, kGridSize);
+            return KThreadSliceSize * (kPerBlock / K_BlockTileSize);
+        }
+    }
+
+    // Calculate mean and variance by welford along k dimension
+    __device__ static void Run(const XGridDesc_M_K& x_grid_desc_m_k,
+                               const MeanVarGridDesc_M_KBlock& mean_var_grid_desc_m_kblock,
+                               index_t num_k_block_tile_iteration,
+                               const XDataType* const __restrict__ p_x_global,
+                               MeanVarDataType* const p_mean_global,
+                               MeanVarDataType* const p_variance_global,
+                               int32_t* const p_welford_count_global)
+    {
+        auto x_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * XSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            var_thread_buf;
+
+        const index_t thread_local_id = get_thread_local_1d_id();
+        const index_t block_global_id = get_block_1d_id();
+
+        const index_t k_grid_size        = mean_var_grid_desc_m_kblock.GetLength(I1);
+        const index_t block_m_cluster_id = block_global_id / k_grid_size;
+        const index_t block_k_cluster_id = block_global_id % k_grid_size;
+
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        const index_t reduceSizePerBlock = K_BlockTileSize * num_k_block_tile_iteration;
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  ComputeDataType,
+                                                                  XGridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  ThreadBufferDimAccessOrder,
+                                                                  XSrcVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(
+                block_m_cluster_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                block_k_cluster_id * reduceSizePerBlock + thread_k_cluster_id * XSrcVectorSize));
+
+        auto mean_var_count_store_index = make_multi_index(
+            block_m_cluster_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+            block_k_cluster_id);
+
+        auto threadwise_welford_mean_var_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               MeanVarDataType,
+                                               decltype(thread_buffer_desc_m_1),
+                                               MeanVarGridDesc_M_KBlock,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M_1,
+                                               Sequence<0, 1>,
+                                               1,
+                                               1,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                mean_var_grid_desc_m_kblock, mean_var_count_store_index, PassThroughOp{});
+
+        constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x_global, x_grid_desc_m_k.GetElementSpaceSize());
+
+        auto mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_mean_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        auto var_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_variance_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        auto threadwise_welford = ThreadwiseWelford();
+        int kRaw                = x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0];
+        threadwise_welford.max_count_ =
+            GetKPerThread(kRaw, k_grid_size, block_k_cluster_id, thread_k_cluster_id);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I) = type_convert<ComputeDataType>(0.0f);
+            var_thread_buf(I)  = type_convert<ComputeDataType>(0.0f);
+        });
+
+        for(index_t k = 0; k < num_k_block_tile_iteration; ++k)
+        {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_x_load.Run(x_grid_desc_m_k,
+                                      x_global_val_buf,
+                                      thread_buffer_desc_m_k,
+                                      make_tuple(I0, I0),
+                                      x_thread_buf(i));
+                threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+                threadwise_welford.Run(x_thread_buf[i], mean_thread_buf, var_thread_buf);
+            });
+        }
+
+        int welford_count = 0;
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            int count = threadwise_welford.cur_count_;
+            BlockwiseWelford::Run(mean_thread_buf(I), var_thread_buf(I), count);
+
+            // The value of count is same for all I
+            if constexpr(I == MThreadSliceSize - 1)
+                welford_count = count;
+        });
+
+        if(thread_k_cluster_id == 0)
+        {
+            threadwise_welford_mean_var_store.Run(thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  mean_thread_buf,
+                                                  mean_var_grid_desc_m_kblock,
+                                                  mean_global_val_buf);
+
+            threadwise_welford_mean_var_store.Run(thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  var_thread_buf,
+                                                  mean_var_grid_desc_m_kblock,
+                                                  var_global_val_buf);
+
+            if(block_m_cluster_id == 0 && thread_m_cluster_id == 0)
+                p_welford_count_global[block_k_cluster_id] = welford_count;
+        }
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..136ac94e7f0fabb81cfd5baea7e58174799f4d43
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
@@ -0,0 +1,418 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/math.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename MeanVarDataType,
+          typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
+          typename MeanVarGridDesc_M_KBlock,
+          typename CountGridDesc_M_KBlock,
+          typename XYGammaBetaGridDesc_M_K,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcVectorDim,
+          index_t XSrcVectorSize,
+          index_t GammaSrcVectorDim,
+          index_t GammaSrcVectorSize,
+          index_t BetaSrcVectorDim,
+          index_t BetaSrcVectorSize,
+          index_t YDstVectorDim,
+          index_t YDstVectorSize>
+struct GridwiseNormalizationSplitK2nd
+{
+    static_assert((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert((YDstVectorDim == 0 && MThreadSliceSize % YDstVectorSize == 0) ||
+                      (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert(XSrcVectorSize == YDstVectorSize);
+    static_assert(XSrcVectorSize == GammaSrcVectorSize);
+    static_assert(XSrcVectorSize == BetaSrcVectorSize);
+
+    static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0);
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using ThreadBufferDimAccessOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadBufferLengths_M_K                = Sequence<MThreadSliceSize, XSrcVectorSize>;
+    static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
+
+    using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
+    static constexpr auto thread_buffer_desc_m_1 =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, I1));
+
+    using ThreadWelfordSrcDesc_M_1 = decltype(thread_buffer_desc_m_1);
+    using ThreadWelfordDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using ThreadwiseWelford =
+        ThreadwiseWelfordMerge<ComputeDataType, ThreadWelfordSrcDesc_M_1, ThreadWelfordDstDesc_M>;
+
+    using BlockwiseWelford = BlockwiseWelford<ComputeDataType,
+                                              BlockSize,
+                                              ThreadClusterLengths_M_K,
+                                              ThreadClusterArrangeOrder>;
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    static constexpr index_t M_BlockTileSize     = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize     = KThreadClusterSize * KThreadSliceSize;
+    static constexpr index_t K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize;
+
+    static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
+
+    __device__ static void Run(const MeanVarGridDesc_M_KBlock& mean_var_grid_desc_m_kblock,
+                               const CountGridDesc_M_KBlock& count_grid_desc_m_kblock,
+                               const XYGammaBetaGridDesc_M_K& x_grid_desc_m_k,
+                               const XYGammaBetaGridDesc_M_K& gamma_grid_desc_m_k,
+                               const XYGammaBetaGridDesc_M_K& beta_grid_desc_m_k,
+                               const XYGammaBetaGridDesc_M_K& y_grid_desc_m_k,
+                               index_t num_k_mean_var_count_iteration,
+                               index_t num_k_block_tile_iteration,
+                               index_t k_grid_size,
+                               ComputeDataType epsilon,
+                               const MeanVarDataType* const p_mean_global,
+                               const MeanVarDataType* const p_variance_global,
+                               const int32_t* const p_welford_count_global,
+                               const XDataType* const __restrict__ p_x_global,
+                               const GammaDataType* const __restrict__ p_gamma_global,
+                               const BetaDataType* const __restrict__ p_beta_global,
+                               YDataType* const __restrict__ p_y_global,
+                               const YElementwiseOperation y_elementwise_op)
+    {
+        // Thread/Block id
+        const index_t thread_local_id    = get_thread_local_1d_id();
+        const index_t block_global_id    = get_block_1d_id();
+        const index_t block_m_cluster_id = block_global_id / k_grid_size;
+        const index_t block_k_cluster_id = block_global_id % k_grid_size;
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        // Global Memory
+        const auto mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_mean_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        const auto var_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_variance_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        const auto welford_count_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_welford_count_global, count_grid_desc_m_kblock.GetElementSpaceSize());
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x_global, x_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto gamma_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_gamma_global, gamma_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto beta_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_beta_global, beta_grid_desc_m_k.GetElementSpaceSize());
+
+        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
+
+        // VGPR
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            in_mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            in_var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true>
+            in_welford_count_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true>
+            welford_count_thread_buf;
+
+        auto x_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * XSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        auto gamma_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * GammaSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        auto& beta_thread_buf = gamma_thread_buf;
+        auto& y_thread_buf    = x_thread_buf;
+
+        // IO
+        auto threadwise_mean_var_load_m_kblock =
+            ThreadwiseTensorSliceTransfer_v2<MeanVarDataType,
+                                             ComputeDataType,
+                                             MeanVarGridDesc_M_KBlock,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             1,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_grid_desc_m_kblock,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id));
+
+        auto threadwise_count_load_m_kblock =
+            ThreadwiseTensorSliceTransfer_v2<int32_t,
+                                             int32_t,
+                                             CountGridDesc_M_KBlock,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             1,
+                                             1,
+                                             1,
+                                             true>(
+                count_grid_desc_m_kblock,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id));
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  ComputeDataType,
+                                                                  XYGammaBetaGridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  ThreadBufferDimAccessOrder,
+                                                                  XSrcVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize,
+                             block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                 thread_k_cluster_id * XSrcVectorSize));
+
+        auto threadwise_gamma_load =
+            ThreadwiseTensorSliceTransfer_v2<GammaDataType,
+                                             ComputeDataType,
+                                             XYGammaBetaGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             ThreadBufferDimAccessOrder,
+                                             GammaSrcVectorDim,
+                                             GammaSrcVectorSize,
+                                             1,
+                                             true>(
+                gamma_grid_desc_m_k,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                     thread_k_cluster_id * GammaSrcVectorSize));
+
+        auto threadwise_beta_load =
+            ThreadwiseTensorSliceTransfer_v2<BetaDataType,
+                                             ComputeDataType,
+                                             XYGammaBetaGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             ThreadBufferDimAccessOrder,
+                                             BetaSrcVectorDim,
+                                             BetaSrcVectorSize,
+                                             1,
+                                             true>(
+                beta_grid_desc_m_k,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                     thread_k_cluster_id * BetaSrcVectorSize));
+
+        auto threadwise_y_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               YDataType,
+                                               decltype(thread_buffer_desc_m_k),
+                                               XYGammaBetaGridDesc_M_K,
+                                               YElementwiseOperation,
+                                               ThreadBufferLengths_M_K,
+                                               ThreadBufferDimAccessOrder,
+                                               YDstVectorDim,
+                                               YDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                y_grid_desc_m_k,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                     thread_k_cluster_id * YDstVectorSize),
+                y_elementwise_op);
+
+        // step1: Merge mean and variance
+        constexpr auto mean_var_count_thread_copy_step_I0_k =
+            make_multi_index(I0, KThreadClusterSize);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I)          = type_convert<ComputeDataType>(0.0f);
+            var_thread_buf(I)           = type_convert<ComputeDataType>(0.0f);
+            welford_count_thread_buf(I) = 0;
+        });
+
+        for(index_t k = 0; k < num_k_mean_var_count_iteration; ++k)
+        {
+            threadwise_mean_var_load_m_kblock.Run(mean_var_grid_desc_m_kblock,
+                                                  mean_global_val_buf,
+                                                  thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  in_mean_thread_buf);
+
+            threadwise_mean_var_load_m_kblock.Run(mean_var_grid_desc_m_kblock,
+                                                  var_global_val_buf,
+                                                  thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  in_var_thread_buf);
+
+            threadwise_count_load_m_kblock.Run(count_grid_desc_m_kblock,
+                                               welford_count_global_val_buf,
+                                               thread_buffer_desc_m_1,
+                                               make_tuple(I0, I0),
+                                               in_welford_count_thread_buf);
+
+            ThreadwiseWelford::Run(in_mean_thread_buf,
+                                   in_var_thread_buf,
+                                   in_welford_count_thread_buf,
+                                   mean_thread_buf,
+                                   var_thread_buf,
+                                   welford_count_thread_buf);
+
+            threadwise_mean_var_load_m_kblock.MoveSrcSliceWindow(
+                mean_var_grid_desc_m_kblock, mean_var_count_thread_copy_step_I0_k);
+            threadwise_count_load_m_kblock.MoveSrcSliceWindow(count_grid_desc_m_kblock,
+                                                              mean_var_count_thread_copy_step_I0_k);
+        }
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            BlockwiseWelford::Run(
+                mean_thread_buf(I), var_thread_buf(I), welford_count_thread_buf(I));
+        });
+
+        // step2: normalization
+        constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
+
+        for(index_t k = 0; k < num_k_block_tile_iteration; ++k)
+        {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_x_load.Run(x_grid_desc_m_k,
+                                      x_global_val_buf,
+                                      thread_buffer_desc_m_k,
+                                      make_tuple(I0, I0),
+                                      x_thread_buf(i));
+                threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_gamma_load.Run(gamma_grid_desc_m_k,
+                                          gamma_global_val_buf,
+                                          thread_buffer_desc_m_k,
+                                          make_tuple(I0, I0),
+                                          gamma_thread_buf(i));
+
+                threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                         thread_copy_fwd_step_m_k);
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                    static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                        constexpr auto offset_m_k =
+                            thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                        // normalize
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
+                            divisor;
+
+                        // gamma
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) *
+                            gamma_thread_buf(iK0)(Number<offset_m_k>{});
+                    });
+                });
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_beta_load.Run(beta_grid_desc_m_k,
+                                         beta_global_val_buf,
+                                         thread_buffer_desc_m_k,
+                                         make_tuple(I0, I0),
+                                         beta_thread_buf(i));
+                threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                        thread_copy_fwd_step_m_k);
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                    static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                        constexpr auto offset_m_k =
+                            thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                        // beta
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) +
+                            beta_thread_buf(iK0)(Number<offset_m_k>{});
+                    });
+                });
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_y_store.Run(thread_buffer_desc_m_k,
+                                       make_tuple(I0, I0),
+                                       y_thread_buf(i),
+                                       y_grid_desc_m_k,
+                                       y_global_val_buf);
+                threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_fwd_step_m_k);
+            });
+        } // end for (normalization)
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
similarity index 61%
rename from include/ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp
rename to include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
index 7aefd3c066b3b29c2af12bfce8cf98b5b900b755..ff9712276c26d14c9a68f9df618ed4666f2dc5e9 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -16,8 +16,8 @@ template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
-          typename AccDataType,
-          typename AccElementwiseOperation,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
           typename GridDesc_M_K,
           index_t BlockSize,
           index_t MThreadClusterSize,
@@ -43,6 +43,10 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                       (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
                   "Invalid thread slice sizes and/or vector sizes configuration, please check!");
 
+    static_assert(XSrcVectorSize == YDstVectorSize);
+    static_assert(XSrcVectorSize == GammaSrcVectorSize);
+    static_assert(XSrcVectorSize == BetaSrcVectorSize);
+
     static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0);
 
     using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
@@ -56,15 +60,19 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
     static constexpr auto thread_cluster_desc =
         make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
 
+    using ThreadBufferLengths_M_K                = Sequence<MThreadSliceSize, XSrcVectorSize>;
+    static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
+
     using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
         make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{})));
     using ThreadReduceDstDesc_M =
         decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
 
     using ThreadwiseWelford =
-        ThreadwiseWelford<AccDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
+        ThreadwiseWelford<ComputeDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
 
-    using BlockwiseWelford = BlockwiseWelford<AccDataType,
+    using BlockwiseWelford = BlockwiseWelford<ComputeDataType,
                                               BlockSize,
                                               ThreadClusterLengths_M_K,
                                               ThreadClusterArrangeOrder>;
@@ -77,10 +85,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
     static constexpr index_t K_BlockTileSize     = KThreadClusterSize * KThreadSliceSize;
     static constexpr index_t K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize;
 
-    static constexpr auto XThreadBufferNumber     = Number<KThreadSliceSize / XSrcVectorSize>{};
-    static constexpr auto GammaThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
-    static constexpr auto BetaThreadBufferNumber  = Number<KThreadSliceSize / XSrcVectorSize>{};
-    static constexpr auto YThreadBufferNumber     = Number<KThreadSliceSize / XSrcVectorSize>{};
+    static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
 
     __device__ static int GetKPerThread(const GridDesc_M_K& x_grid_desc_m_k,
                                         int thread_k_cluster_id)
@@ -93,7 +98,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
 
         if(kPerBlockTail > 0)
         {
-            static_for<0, XThreadBufferNumber, 1>{}([&](auto i) {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
                 int thread_max_len =
                     (thread_k_cluster_id + 1) * XSrcVectorSize + K_BlockTileStepSize * i;
                 int delta = thread_max_len - kPerBlockTail;
@@ -110,59 +115,41 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                                const GridDesc_M_K& beta_grid_desc_m_k,
                                const GridDesc_M_K& y_grid_desc_m_k,
                                index_t num_k_block_tile_iteration,
-                               AccDataType epsilon,
+                               ComputeDataType epsilon,
                                const XDataType* const __restrict__ p_x_global,
                                const GammaDataType* const __restrict__ p_gamma_global,
                                const BetaDataType* const __restrict__ p_beta_global,
                                YDataType* const __restrict__ p_y_global,
-                               const AccElementwiseOperation acc_elementwise_op)
+                               const YElementwiseOperation y_elementwise_op)
     {
-        if constexpr(SweepOnce)
-        {
-            num_k_block_tile_iteration = 1;
-        }
-
         auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
 
         auto x_thread_buf = generate_tuple(
             [&](auto) {
                 return StaticBuffer<AddressSpaceEnum::Vgpr,
-                                    AccDataType,
+                                    ComputeDataType,
                                     MThreadSliceSize * XSrcVectorSize,
                                     true>{};
             },
-            Number<XThreadBufferNumber>{});
+            Number<ThreadBufferNumber>{});
 
         auto gamma_thread_buf = generate_tuple(
             [&](auto) {
                 return StaticBuffer<AddressSpaceEnum::Vgpr,
-                                    AccDataType,
+                                    ComputeDataType,
                                     MThreadSliceSize * GammaSrcVectorSize,
                                     true>{};
             },
-            Number<GammaThreadBufferNumber>{});
-
-        auto beta_thread_buf = generate_tuple(
-            [&](auto) {
-                return StaticBuffer<AddressSpaceEnum::Vgpr,
-                                    AccDataType,
-                                    MThreadSliceSize * BetaSrcVectorSize,
-                                    true>{};
-            },
-            Number<BetaThreadBufferNumber>{});
+            Number<ThreadBufferNumber>{});
 
-        auto y_thread_buf = generate_tuple(
-            [&](auto) {
-                return StaticBuffer<AddressSpaceEnum::Vgpr,
-                                    AccDataType,
-                                    MThreadSliceSize * YDstVectorSize,
-                                    true>{};
-            },
-            Number<YThreadBufferNumber>{});
+        auto& beta_thread_buf = gamma_thread_buf;
+        auto& y_thread_buf    = x_thread_buf;
 
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> mean_thread_buf;
-        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            var_thread_buf;
 
         const index_t thread_local_id = get_thread_local_1d_id();
         const index_t block_global_id = get_block_1d_id();
@@ -173,12 +160,8 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
         const auto thread_m_cluster_id = thread_cluster_idx[I0];
         const auto thread_k_cluster_id = thread_cluster_idx[I1];
 
-        using ThreadBufferLengths_M_K         = Sequence<MThreadSliceSize, XSrcVectorSize>;
-        constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
-
         auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
-                                                                  AccDataType,
+                                                                  ComputeDataType,
                                                                   GridDesc_M_K,
                                                                   decltype(thread_buffer_desc_m_k),
                                                                   ThreadBufferLengths_M_K,
@@ -194,7 +177,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
 
         auto threadwise_gamma_load =
             ThreadwiseTensorSliceTransfer_v2<GammaDataType,
-                                             AccDataType,
+                                             ComputeDataType,
                                              GridDesc_M_K,
                                              decltype(thread_buffer_desc_m_k),
                                              ThreadBufferLengths_M_K,
@@ -210,7 +193,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
 
         auto threadwise_beta_load =
             ThreadwiseTensorSliceTransfer_v2<BetaDataType,
-                                             AccDataType,
+                                             ComputeDataType,
                                              GridDesc_M_K,
                                              decltype(thread_buffer_desc_m_k),
                                              ThreadBufferLengths_M_K,
@@ -225,11 +208,11 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                                  thread_k_cluster_id * BetaSrcVectorSize));
 
         auto threadwise_y_store =
-            ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
                                                YDataType,
                                                decltype(thread_buffer_desc_m_k),
                                                GridDesc_M_K,
-                                               AccElementwiseOperation,
+                                               YElementwiseOperation,
                                                ThreadBufferLengths_M_K,
                                                ThreadBufferDimAccessOrder,
                                                YDstVectorDim,
@@ -241,7 +224,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                 make_multi_index(block_global_id * M_BlockTileSize +
                                      thread_m_cluster_id * MThreadSliceSize,
                                  thread_k_cluster_id * YDstVectorSize),
-                acc_elementwise_op);
+                y_elementwise_op);
 
         constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
         constexpr auto thread_copy_bwd_step_m_k =
@@ -260,67 +243,47 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
         threadwise_welford.max_count_ = GetKPerThread(x_grid_desc_m_k, thread_k_cluster_id);
 
         static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
-            mean_thread_buf(I) = type_convert<AccDataType>(0.0f);
-            var_thread_buf(I)  = type_convert<AccDataType>(0.0f);
+            mean_thread_buf(I) = type_convert<ComputeDataType>(0.0f);
+            var_thread_buf(I)  = type_convert<ComputeDataType>(0.0f);
         });
 
-        for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+        // Separate sweep once and sweep twice pipeline
+        if constexpr(SweepOnce)
         {
-            static_for<0, XThreadBufferNumber, 1>{}([&](auto i) {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
                 threadwise_x_load.Run(x_grid_desc_m_k,
                                       x_global_val_buf,
                                       thread_buffer_desc_m_k,
                                       make_tuple(I0, I0),
                                       x_thread_buf(i));
-                threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
-                threadwise_welford.Run(x_thread_buf[i], mean_thread_buf, var_thread_buf);
-            });
-        }
-
-        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
-            if constexpr(I > 0)
-                block_sync_lds();
 
-            int count = threadwise_welford.cur_count_;
-            BlockwiseWelford::Run(mean_thread_buf(I), var_thread_buf(I), count);
-        });
-
-        auto thread_copy_tail_m_k =
-            (num_k_block_tile_iteration - 1) * XThreadBufferNumber * thread_copy_fwd_step_m_k;
-
-        threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_bwd_step_m_k);
-        threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k, thread_copy_tail_m_k);
-        threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k, thread_copy_tail_m_k);
-        threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_tail_m_k);
-
-        for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
-        {
-            if constexpr(!SweepOnce)
-            {
-                static_for<0, XThreadBufferNumber, 1>{}([&](auto i) {
-                    threadwise_x_load.Run(x_grid_desc_m_k,
-                                          x_global_val_buf,
-                                          thread_buffer_desc_m_k,
-                                          make_tuple(I0, I0),
-                                          x_thread_buf(i));
-                    threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
-                });
-            }
-
-            static_for<0, GammaThreadBufferNumber, 1>{}([&](auto i) {
                 threadwise_gamma_load.Run(gamma_grid_desc_m_k,
                                           gamma_global_val_buf,
                                           thread_buffer_desc_m_k,
                                           make_tuple(I0, I0),
                                           gamma_thread_buf(i));
 
-                threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
-                                                         thread_copy_fwd_step_m_k);
+                threadwise_welford.Run(x_thread_buf[i], mean_thread_buf, var_thread_buf);
+
+                if constexpr(i != ThreadBufferNumber - 1)
+                {
+                    threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+                    threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                             thread_copy_fwd_step_m_k);
+                }
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                if constexpr(I > 0)
+                    block_sync_lds();
+
+                int count = threadwise_welford.cur_count_;
+                BlockwiseWelford::Run(mean_thread_buf(I), var_thread_buf(I), count);
             });
 
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                auto divisor = 1 / __builtin_amdgcn_sqrtf(var_thread_buf(iM) + epsilon);
-                static_for<0, XThreadBufferNumber, 1>{}([&](auto iK0) {
+                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =
                             thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
@@ -330,7 +293,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                             (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
                             divisor;
 
-                        // gamma
+                        // gamma & beta
                         y_thread_buf(iK0)(Number<offset_m_k>{}) =
                             y_thread_buf(iK0)(Number<offset_m_k>{}) *
                             gamma_thread_buf(iK0)(Number<offset_m_k>{});
@@ -338,18 +301,20 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                 });
             });
 
-            static_for<0, BetaThreadBufferNumber, 1>{}([&](auto i) {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
                 threadwise_beta_load.Run(beta_grid_desc_m_k,
                                          beta_global_val_buf,
                                          thread_buffer_desc_m_k,
                                          make_tuple(I0, I0),
                                          beta_thread_buf(i));
-                threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
-                                                        thread_copy_fwd_step_m_k);
+
+                if constexpr(i != ThreadBufferNumber - 1)
+                    threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                            thread_copy_fwd_step_m_k);
             });
 
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                static_for<0, XThreadBufferNumber, 1>{}([&](auto iK0) {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =
                             thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
@@ -362,22 +327,134 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                 });
             });
 
-            static_for<0, YThreadBufferNumber, 1>{}([&](auto i) {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
                 threadwise_y_store.Run(thread_buffer_desc_m_k,
                                        make_tuple(I0, I0),
                                        y_thread_buf(i),
                                        y_grid_desc_m_k,
                                        y_global_val_buf);
-                threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_fwd_step_m_k);
+
+                if constexpr(i != ThreadBufferNumber - 1)
+                    threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k,
+                                                          thread_copy_fwd_step_m_k);
+            });
+        } // end of sweep once
+        else
+        {
+            for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+            {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_x_load.Run(x_grid_desc_m_k,
+                                          x_global_val_buf,
+                                          thread_buffer_desc_m_k,
+                                          make_tuple(I0, I0),
+                                          x_thread_buf(i));
+                    threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+                    threadwise_welford.Run(x_thread_buf[i], mean_thread_buf, var_thread_buf);
+                });
+            }
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                if constexpr(I > 0)
+                    block_sync_lds();
+
+                int count = threadwise_welford.cur_count_;
+                BlockwiseWelford::Run(mean_thread_buf(I), var_thread_buf(I), count);
             });
 
-            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, 2 * thread_copy_bwd_step_m_k);
-            threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
-                                                     2 * thread_copy_bwd_step_m_k);
-            threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
-                                                    2 * thread_copy_bwd_step_m_k);
-            threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, 2 * thread_copy_bwd_step_m_k);
-        }
+            auto thread_copy_tail_m_k =
+                (num_k_block_tile_iteration - 1) * ThreadBufferNumber * thread_copy_fwd_step_m_k;
+
+            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_bwd_step_m_k);
+            threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k, thread_copy_tail_m_k);
+            threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k, thread_copy_tail_m_k);
+            threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_tail_m_k);
+
+            for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+            {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_x_load.Run(x_grid_desc_m_k,
+                                          x_global_val_buf,
+                                          thread_buffer_desc_m_k,
+                                          make_tuple(I0, I0),
+                                          x_thread_buf(i));
+                    threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+                });
+
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_gamma_load.Run(gamma_grid_desc_m_k,
+                                              gamma_global_val_buf,
+                                              thread_buffer_desc_m_k,
+                                              make_tuple(I0, I0),
+                                              gamma_thread_buf(i));
+
+                    threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                             thread_copy_fwd_step_m_k);
+                });
+
+                static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                    auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
+                    static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                        static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                            constexpr auto offset_m_k =
+                                thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                            // normalize
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                                (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
+                                divisor;
+
+                            // gamma
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                                y_thread_buf(iK0)(Number<offset_m_k>{}) *
+                                gamma_thread_buf(iK0)(Number<offset_m_k>{});
+                        });
+                    });
+                });
+
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_beta_load.Run(beta_grid_desc_m_k,
+                                             beta_global_val_buf,
+                                             thread_buffer_desc_m_k,
+                                             make_tuple(I0, I0),
+                                             beta_thread_buf(i));
+                    threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                            thread_copy_fwd_step_m_k);
+                });
+
+                static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                    static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                        static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                            constexpr auto offset_m_k =
+                                thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                            // beta
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                                y_thread_buf(iK0)(Number<offset_m_k>{}) +
+                                beta_thread_buf(iK0)(Number<offset_m_k>{});
+                        });
+                    });
+                });
+
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    threadwise_y_store.Run(thread_buffer_desc_m_k,
+                                           make_tuple(I0, I0),
+                                           y_thread_buf(i),
+                                           y_grid_desc_m_k,
+                                           y_global_val_buf);
+                    threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k,
+                                                          thread_copy_fwd_step_m_k);
+                });
+
+                threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, 2 * thread_copy_bwd_step_m_k);
+                threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                         2 * thread_copy_bwd_step_m_k);
+                threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                        2 * thread_copy_bwd_step_m_k);
+                threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k,
+                                                      2 * thread_copy_bwd_step_m_k);
+            }
+        } // end of sweep twice
     }
 };
 
diff --git a/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp b/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp
index 188c62d93b0d60317ac139115b79f496612d2ea6..c6eecc067d9f77fa488e7a922fe1fb711c0c0d0f 100644
--- a/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp
index 94cdfe01087bc4005d45f11992a6a8f94d42691c..44730d551c5c8c9fb2408502ece0d45c6f37a221 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp
index e045e3b545a9897f5df9df1948123e86424d7c7b..e97aa433a6f3b9d54fe76ed26b5b0f6bd439ae84 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_THREADWISE_GEMM_DLOPS_V3_HPP
 #define CK_THREADWISE_GEMM_DLOPS_V3_HPP
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp
index 0a1197a1630c12c53afa9513f10fcfa907ba82d3..6774a35bcb872466d272da0bccd618709d3fb29b 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
index b0f453b025f00803091b200cc7201e47e1d4993e..605f2569c6622bb589d46da71020ea46ddba69f7 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -1201,7 +1201,12 @@ struct ThreadwiseTensorSliceTransfer_v4
     SrcCoord src_ref_coord_;
 };
 
-// Do NOT involve any tensor coordinates with StaticBuffer
+/**
+ * @brief Threadwise data transfer
+ *
+ * Do NOT involve any tensor coordinates with StaticBuffer
+ *
+ */
 template <typename SrcData,
           typename DstData,
           typename SrcDesc,
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
index bb28c194f4b7a2f371d2be00cb00fc4fe3c17d3f..6665d765f81ccd45f36598b104711e77368b6ee5 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
@@ -1,11 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck/utility/common_header.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor/static_tensor.hpp"
 
@@ -207,15 +208,6 @@ struct ThreadwiseTensorSliceTransfer_v3r1
             auto src_vector_container = src_vector_type{
                 src_buf.template Get<src_vector_t>(src_coord_.GetOffset(), is_src_valid)};
 
-            // apply SrcElementwiseOperation on src_vector_container
-            static_for<0, SrcScalarPerVector, 1>{}([&](auto i) {
-                SrcData src_v;
-
-                src_element_op_(src_v, src_vector_container.template AsType<SrcData>()[i]);
-
-                src_vector_container.template AsType<SrcData>()(i) = src_v;
-            });
-
             // copy data from src_vector_container into src_thread_scratch_
             src_thread_scratch_tuple_(thread_scratch_id)
                 .template SetAsType<src_vector_t>(
@@ -318,7 +310,6 @@ struct ThreadwiseTensorSliceTransfer_v3r1
                 constexpr auto data_idx_seq = generate_sequence_v2(
                     [&](auto i) { return Number<data_idx[i]>{}; }, Number<nDim>{});
 
-                // TODO type_convert is not used yet!!!!!
                 using src_vector_t = vector_type_maker_t<SrcData, SrcScalarPerVector>;
                 using dst_vector_t = vector_type_maker_t<DstData, DstScalarPerVector>;
 
@@ -342,19 +333,17 @@ struct ThreadwiseTensorSliceTransfer_v3r1
                     Number<num_dst_vector>{});
 
                 // do data transpose
-                // TODO type_convert is not used yet!!!!!
                 transpose_vectors<SrcData, DstScalarPerVector, SrcScalarPerVector>{}(
                     src_vector_refs, dst_vector_refs);
             });
         }
-        else
-        {
-            static_ford<SliceLengths>{}([&](auto idx) {
-                // convert from SrcData to DstData here
-                dst_thread_scratch_(idx) =
-                    type_convert<DstData>(src_thread_scratch_tuple_[thread_scratch_id][idx]);
-            });
-        }
+
+        static_ford<SliceLengths>{}([&](auto idx) {
+            // apply the src elementwise op and convert to DstData under the hood if needed
+            DstData dst_v;
+            src_element_op_(dst_v, src_thread_scratch_tuple_[thread_scratch_id][idx]);
+            dst_thread_scratch_(idx) = dst_v;
+        });
 #endif
     }
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r3.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r3.hpp
deleted file mode 100644
index 6a73466efa443af294f2062ee5173cf548880c3d..0000000000000000000000000000000000000000
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r3.hpp
+++ /dev/null
@@ -1,886 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_THREADWISE_TENSOR_SLICE_TRANSFER_V3R3_HPP
-#define CK_THREADWISE_TENSOR_SLICE_TRANSFER_V3R3_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "static_tensor.hpp"
-
-namespace ck {
-
-namespace detail {
-// TODO: How to fix this? It uses an struct instead of lambda because lambda
-// doesn't have constructor
-template <index_t SrcVectorDim,
-          index_t SrcScalarPerVector,
-          index_t DstVectorDim,
-          index_t DstScalarPerVector>
-struct lambda_scalar_per_access_for_src_and_dst
-{
-    __host__ __device__ constexpr auto operator()(index_t i) const
-    {
-        if(i == SrcVectorDim && i == DstVectorDim)
-        {
-            return math::lcm(SrcScalarPerVector, DstScalarPerVector);
-        }
-        else if(i == SrcVectorDim)
-        {
-            return SrcScalarPerVector;
-        }
-        else if(i == DstVectorDim)
-        {
-            return DstScalarPerVector;
-        }
-        else
-        {
-            return 1;
-        }
-    }
-};
-
-} // namespace detail
-
-// Assume:
-//   1. src_desc and dst_desc are not known at compile-time
-//   2. SrcBuffer and DstBuffer are DynamicBuffer
-//   3. src_slice_origin and dst_slice_origin are not known at compile-time,
-//   4. Use thread buffer
-template <typename SliceLengths,
-          typename SrcElementwiseOperation,
-          typename DstElementwiseOperation,
-          InMemoryDataOperationEnum DstInMemOp,
-          typename SrcData,
-          typename DstData,
-          typename SrcDesc,
-          typename DstDesc,
-          typename Dst0Desc,
-          typename Dst1Desc,
-          typename SrcDimAccessOrder,
-          typename DstDimAccessOrder,
-          index_t SrcVectorDim,
-          index_t DstVectorDim,
-          index_t SrcScalarPerVector,
-          index_t DstScalarPerVector,
-          index_t SrcScalarStrideInVector,
-          index_t DstScalarStrideInVector,
-          bool SrcResetCoordinateAfterRun, // control whether to move back src coordinate after each
-                                           // RunRead(),  will be fused with MoveSrcSliceWindow to
-                                           // save addr computation
-          bool DstResetCoordinateAfterRun> // control whether to move back dst coordinate after each
-                                           // RunWrite(),  will be fused with MoveDstSliceWindow to
-                                           // save addr computation
-struct ThreadwiseTensorSliceTransfer_v3r3
-{
-    static constexpr index_t nDim = SliceLengths::Size();
-    using Index                   = MultiIndex<nDim>;
-
-    using SrcCoord  = decltype(make_tensor_coordinate(SrcDesc{}, Index{}));
-    using DstCoord  = decltype(make_tensor_coordinate(DstDesc{}, Index{}));
-    using Dst0Coord = decltype(make_tensor_coordinate(Dst0Desc{}, Index{}));
-    using Dst1Coord = decltype(make_tensor_coordinate(Dst1Desc{}, Index{}));
-
-    using SrcCoordStep  = decltype(make_tensor_coordinate_step(SrcDesc{}, Index{}));
-    using DstCoordStep  = decltype(make_tensor_coordinate_step(DstDesc{}, Index{}));
-    using Dst0CoordStep = decltype(make_tensor_coordinate_step(Dst0Desc{}, Index{}));
-    using Dst1CoordStep = decltype(make_tensor_coordinate_step(Dst1Desc{}, Index{}));
-
-    __device__ constexpr ThreadwiseTensorSliceTransfer_v3r3(
-        const SrcDesc& src_desc,
-        const Index& src_slice_origin,
-        const SrcElementwiseOperation& src_element_op,
-        const DstDesc& dst_desc,
-        const Dst0Desc& dst0_desc,
-        const Dst1Desc& dst1_desc,
-        const Index& dst_slice_origin,
-        const DstElementwiseOperation& dst_element_op)
-        : src_coord_(make_tensor_coordinate(src_desc, src_slice_origin)),
-          dst_coord_(make_tensor_coordinate(dst_desc, dst_slice_origin)),
-          dst0_coord_(make_tensor_coordinate(dst0_desc, dst_slice_origin)),
-          dst1_coord_(make_tensor_coordinate(dst1_desc, dst_slice_origin)),
-          src_element_op_(src_element_op),
-          dst_element_op_(dst_element_op)
-    {
-    }
-
-    __device__ void SetSrcSliceOrigin(const SrcDesc& src_desc, const Index& src_slice_origin_idx)
-    {
-        src_coord_ = make_tensor_coordinate(src_desc, src_slice_origin_idx);
-    }
-
-    __device__ void SetDstSliceOrigin(const DstDesc& dst_desc,
-                                      const Dst0Desc& dst0_desc,
-                                      const Dst1Desc& dst1_desc,
-                                      const Index& dst_slice_origin_idx)
-    {
-        dst_coord_  = make_tensor_coordinate(dst_desc, dst_slice_origin_idx);
-        dst0_coord_ = make_tensor_coordinate(dst0_desc, dst_slice_origin_idx);
-        dst1_coord_ = make_tensor_coordinate(dst1_desc, dst_slice_origin_idx);
-    }
-
-    template <typename SrcBuffer>
-    __device__ void RunRead(const SrcDesc& src_desc, const SrcBuffer& src_buf)
-    {
-        static_assert(SrcBuffer::GetAddressSpace() == AddressSpaceEnum::Global or
-                          SrcBuffer::GetAddressSpace() == AddressSpaceEnum::Lds,
-                      "wrong!");
-
-        static_assert(
-            is_same<remove_cvref_t<typename SrcBuffer::type>, remove_cvref_t<SrcData>>::value,
-            "wrong! SrcBuffer and SrcData data type are inconsistent");
-
-        constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-
-        // scalar per access on each dim
-        // TODO: don't use lambda_scalar_per_access
-        constexpr auto src_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<SrcVectorDim, SrcScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto src_access_lengths = SliceLengths{} / src_scalar_per_access;
-
-        constexpr auto src_dim_access_order = SrcDimAccessOrder{};
-
-        constexpr auto ordered_src_access_lengths =
-            container_reorder_given_new2old(src_access_lengths, src_dim_access_order);
-
-        // make forward steps
-        const auto src_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? src_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(src_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps
-        const auto src_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -src_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(src_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // loop over tensor and copy
-        static_ford<decltype(ordered_src_access_lengths)>{}([&](auto ordered_src_access_idx) {
-            // judge move forward or move backward
-            constexpr auto forward_sweep = [&]() {
-                StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-                forward_sweep_(I0) = true;
-
-                static_for<1, nDim, 1>{}([&](auto i) {
-                    index_t tmp = ordered_src_access_idx[I0];
-
-                    static_for<1, i, 1>{}([&](auto j) {
-                        tmp = tmp * ordered_src_access_lengths[j] + ordered_src_access_idx[j];
-                    });
-
-                    forward_sweep_(i) = tmp % 2 == 0;
-                });
-
-                return forward_sweep_;
-            }();
-
-            // calculate src data index
-            constexpr auto src_data_idx = [&]() {
-                Index ordered_idx;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    ordered_idx(i) = forward_sweep[i] ? ordered_src_access_idx[i]
-                                                      : ordered_src_access_lengths[i] - 1 -
-                                                            ordered_src_access_idx[i];
-                });
-
-                return container_reorder_given_old2new(ordered_idx, src_dim_access_order) *
-                       src_scalar_per_access;
-            }();
-
-            constexpr auto src_data_idx_seq = generate_sequence_v2(
-                [&](auto i) { return Number<src_data_idx[i]>{}; }, Number<src_data_idx.Size()>{});
-
-            const bool is_src_valid =
-                coordinate_has_valid_offset_assuming_visible_index_is_valid(src_desc, src_coord_);
-
-            using src_vector_type = vector_type_maker_t<SrcData, SrcScalarPerVector>;
-            using src_vector_t    = typename src_vector_type::type;
-
-            // copy data from src_buf into src_vector_container
-            auto src_vector_container = src_vector_type{
-                src_buf.template Get<src_vector_t>(src_coord_.GetOffset(), is_src_valid)};
-
-            // apply SrcElementwiseOperation on src_vector_container
-            static_for<0, SrcScalarPerVector, 1>{}([&](auto i) {
-                src_vector_container.template AsType<SrcData>()(i) =
-                    src_element_op_(src_vector_container.template AsType<SrcData>()[i]);
-            });
-
-            // copy data from src_vector_container into src_thread_scratch_
-            src_thread_scratch_.template SetAsType<src_vector_t>(
-                src_data_idx_seq, src_vector_container.template AsType<src_vector_t>()[I0]);
-
-            constexpr auto move_on_dim = [&]() constexpr
-            {
-                StaticallyIndexedArray<bool, nDim> move_on_dim_;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    move_on_dim_(i) = ordered_src_access_idx[i] < ordered_src_access_lengths[i] - 1;
-
-                    static_for<i + 1, nDim, 1>{}([&](auto j) {
-                        move_on_dim_(i) &=
-                            ordered_src_access_idx[j] == ordered_src_access_lengths[j] - 1;
-                    });
-                });
-
-                return move_on_dim_;
-            }
-            ();
-
-            // move src coord
-            static_for<0, nDim, 1>{}([&](auto i) {
-                if constexpr(move_on_dim[i])
-                {
-                    if constexpr(forward_sweep[i])
-                    {
-                        move_tensor_coordinate(
-                            src_desc, src_coord_, src_forward_steps[src_dim_access_order[i]]);
-                    }
-                    else
-                    {
-                        move_tensor_coordinate(
-                            src_desc, src_coord_, src_backward_steps[src_dim_access_order[i]]);
-                    }
-                }
-            });
-        });
-
-        // move src coordinate back to slice origin (or not)
-        if constexpr(SrcResetCoordinateAfterRun)
-        {
-            const auto src_reset_step =
-                make_tensor_coordinate_step(src_desc, GetSrcCoordinateResetStep());
-
-            move_tensor_coordinate(src_desc, src_coord_, src_reset_step);
-        }
-    }
-
-    __device__ void TransferDataFromSrcThreadScratchToDstThreadScratch()
-    {
-#if !CK_EXPERIMENTAL_USE_IN_REGISTER_SUB_DWORD_TRANSPOSE
-        static_ford<SliceLengths>{}([&](auto idx) {
-            // convert from SrcData to DstData here
-            dst_thread_scratch_(idx) = type_convert<DstData>(src_thread_scratch_[idx]);
-        });
-#else
-        // sub-dword transpose between src_thread_scratch_ and dst_thread_scratch_
-        // TODO make this logic more generic for more sub-dword datatype
-        if constexpr(SrcVectorDim != DstVectorDim &&
-                     is_same<half_t, remove_cvref_t<SrcData>>::value &&
-                     is_same<half_t, remove_cvref_t<DstData>>::value &&
-                     SrcScalarPerVector % 2 == 0 && DstScalarPerVector % 2 == 0)
-        {
-            // each transpose does
-            // DstScalarPerVector # of src vectors in src_thread_scratch_
-            // SrcScalarPerVector # of dst vectors in dst_thread_scratch_
-            constexpr index_t num_src_vector = Number<DstScalarPerVector>{};
-            constexpr index_t num_dst_vector = Number<SrcScalarPerVector>{};
-
-            // Assume SrcVectorDim is not the same as DstVectorDim, so we do transpose
-            // TODO: make this logic generic for all scenario
-            static_assert(SrcVectorDim != DstVectorDim, "wrong");
-
-            constexpr auto src_scalar_step_in_vector = generate_sequence(
-                detail::lambda_scalar_step_in_vector<SrcVectorDim>{}, Number<nDim>{});
-
-            constexpr auto dst_scalar_step_in_vector = generate_sequence(
-                detail::lambda_scalar_step_in_vector<DstVectorDim>{}, Number<nDim>{});
-
-            constexpr auto scalar_per_access = generate_sequence(
-                detail::lambda_scalar_per_access_for_src_and_dst<SrcVectorDim,
-                                                                 SrcScalarPerVector,
-                                                                 DstVectorDim,
-                                                                 DstScalarPerVector>{},
-                Number<nDim>{});
-
-            constexpr auto access_lengths = SliceLengths{} / scalar_per_access;
-
-            static_ford<decltype(access_lengths)>{}([&](auto access_idx) {
-                constexpr auto data_idx = access_idx * scalar_per_access;
-
-                constexpr auto data_idx_seq = generate_sequence_v2(
-                    [&](auto i) { return Number<data_idx[i]>{}; }, Number<nDim>{});
-
-                // TODO type_convert is not used yet!!!!!
-                using src_vector_t = vector_type_maker_t<SrcData, SrcScalarPerVector>;
-                using dst_vector_t = vector_type_maker_t<DstData, DstScalarPerVector>;
-
-                // get DstScalarPerVector # of read-only references to src vectors from
-                // src_thread_scratch_
-                const auto src_vector_refs = generate_tie(
-                    [&](auto i) -> const src_vector_t& {
-                        // i increment corresponds to movement in DstVectorDim
-                        return src_thread_scratch_.GetVectorTypeReference(
-                            data_idx_seq + i * dst_scalar_step_in_vector);
-                    },
-                    Number<num_src_vector>{});
-
-                // get SrcScalarPerVector # of references to dst vectors from dst_thread_scratch_
-                auto dst_vector_refs = generate_tie(
-                    [&](auto i) -> dst_vector_t& {
-                        // i increment corresponds to movement in SrcVectorDim
-                        return dst_thread_scratch_.GetVectorTypeReference(
-                            data_idx_seq + i * src_scalar_step_in_vector);
-                    },
-                    Number<num_dst_vector>{});
-
-                // do data transpose
-                // TODO type_convert is not used yet!!!!!
-                transpose_vectors<SrcData, DstScalarPerVector, SrcScalarPerVector>{}(
-                    src_vector_refs, dst_vector_refs);
-            });
-        }
-        else
-        {
-            static_ford<SliceLengths>{}([&](auto idx) {
-                // convert from SrcData to DstData here
-                dst_thread_scratch_(idx) = type_convert<DstData>(src_thread_scratch_[idx]);
-            });
-        }
-#endif
-    }
-
-    template <typename DstBuffer, typename Dst0Buffer, typename Dst1Buffer>
-    __device__ void RunWrite(const DstDesc& dst_desc,
-                             DstBuffer& dst_buf,
-                             const Dst0Desc& dst0_desc,
-                             const Dst0Buffer& dst0_buf,
-                             const Dst1Desc& dst1_desc,
-                             const Dst1Buffer& dst1_buf)
-    {
-        // if there is transpose, it's done here
-        // TODO move this elsewhere
-        TransferDataFromSrcThreadScratchToDstThreadScratch();
-
-        static_assert(DstBuffer::GetAddressSpace() == AddressSpaceEnum::Global or
-                          DstBuffer::GetAddressSpace() == AddressSpaceEnum::Lds,
-                      "wrong!");
-
-        static_assert(
-            is_same<remove_cvref_t<typename DstBuffer::type>, remove_cvref_t<DstData>>::value,
-            "wrong! SrcBuffer or DstBuffer data type is wrong");
-
-        constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-
-        // src scalar per access on each dim
-        // TODO: don't use this
-        constexpr auto dst_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<DstVectorDim, DstScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto dst_access_lengths = SliceLengths{} / dst_scalar_per_access;
-
-        constexpr auto dst_dim_access_order = DstDimAccessOrder{};
-
-        constexpr auto ordered_dst_access_lengths =
-            container_reorder_given_new2old(dst_access_lengths, dst_dim_access_order);
-
-        // make forward steps
-        const auto dst_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make forward steps: dst0
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst0_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst0_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make forward steps: dst1
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst1_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst1_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps
-        const auto dst_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps: dst0
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst0_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst0_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps: dst1
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst1_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst1_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // loop over tensor and copy
-        static_ford<decltype(ordered_dst_access_lengths)>{}([&](auto ordered_dst_access_idx) {
-            // judge move forward or move backward
-            constexpr auto forward_sweep = [&]() {
-                StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-                forward_sweep_(I0) = true;
-
-                static_for<1, nDim, 1>{}([&](auto i) {
-                    index_t tmp = ordered_dst_access_idx[I0];
-
-                    static_for<1, i, 1>{}([&](auto j) {
-                        tmp = tmp * ordered_dst_access_lengths[j] + ordered_dst_access_idx[j];
-                    });
-
-                    forward_sweep_(i) = tmp % 2 == 0;
-                });
-
-                return forward_sweep_;
-            }();
-
-            // calculate dst data index
-            constexpr auto dst_data_idx = [&]() {
-                Index ordered_idx;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    ordered_idx(i) = forward_sweep[i] ? ordered_dst_access_idx[i]
-                                                      : ordered_dst_access_lengths[i] - 1 -
-                                                            ordered_dst_access_idx[i];
-                });
-
-                return container_reorder_given_old2new(ordered_idx, dst_dim_access_order) *
-                       dst_scalar_per_access;
-            }();
-
-            constexpr auto dst_data_idx_seq = generate_sequence_v2(
-                [&](auto i) { return Number<dst_data_idx[i]>{}; }, Number<dst_data_idx.Size()>{});
-
-            const bool is_dst_valid =
-                coordinate_has_valid_offset_assuming_visible_index_is_valid(dst_desc, dst_coord_);
-
-            using dst_vector_type = vector_type_maker_t<DstData, DstScalarPerVector>;
-            using dst_vector_t    = typename dst_vector_type::type;
-
-            // copy data from dst_thread_scratch_ into dst_vector_container
-            auto dst_vector_container = dst_vector_type{
-                dst_thread_scratch_.template GetAsType<dst_vector_t>(dst_data_idx_seq)};
-
-            // apply DstElementwiseOperation on dst_vector_container
-            static_for<0, DstScalarPerVector, 1>{}([&](auto i) {
-                dst_vector_container.template AsType<DstData>()(i) =
-                    dst_element_op_(dst_vector_container.template AsType<DstData>()[i]);
-            });
-
-            // copy data from dst_vector_container to dst_buf
-            dst_buf.template Set<dst_vector_t>(
-                dst_coord_.GetOffset(),
-                is_dst_valid,
-                dst_vector_container.template AsType<dst_vector_t>()[I0]);
-
-            constexpr auto move_on_dim = [&]() constexpr
-            {
-                StaticallyIndexedArray<bool, nDim> move_on_dim_;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    move_on_dim_(i) = ordered_dst_access_idx[i] < ordered_dst_access_lengths[i] - 1;
-
-                    static_for<i + 1, nDim, 1>{}([&](auto j) {
-                        move_on_dim_(i) &=
-                            ordered_dst_access_idx[j] == ordered_dst_access_lengths[j] - 1;
-                    });
-                });
-
-                return move_on_dim_;
-            }
-            ();
-
-            // move dst coord
-            static_for<0, nDim, 1>{}([&](auto i) {
-                if constexpr(move_on_dim[i])
-                {
-                    if constexpr(forward_sweep[i])
-                    {
-                        move_tensor_coordinate(
-                            dst_desc, dst_coord_, dst_forward_steps[dst_dim_access_order[i]]);
-                    }
-                    else
-                    {
-                        move_tensor_coordinate(
-                            dst_desc, dst_coord_, dst_backward_steps[dst_dim_access_order[i]]);
-                    }
-                }
-            });
-        });
-
-        // move dst coordinate back to slice origin (or not)
-        if constexpr(DstResetCoordinateAfterRun)
-        {
-            const auto dst_reset_step =
-                make_tensor_coordinate_step(dst_desc, GetDstCoordinateResetStep());
-
-            move_tensor_coordinate(dst_desc, dst_coord_, dst_reset_step);
-        }
-    }
-
-    __device__ static constexpr auto GetSrcCoordinateResetStep()
-    {
-        constexpr auto I0 = Number<0>{};
-
-        // scalar per access on each dim
-        // TODO: don't use lambda_scalar_per_access
-        constexpr auto src_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<SrcVectorDim, SrcScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto src_access_lengths = SliceLengths{} / src_scalar_per_access;
-
-        constexpr auto src_dim_access_order = SrcDimAccessOrder{};
-
-        constexpr auto ordered_src_access_lengths =
-            container_reorder_given_new2old(src_access_lengths, src_dim_access_order);
-
-        // judge move forward or move backward during the last iteration
-        constexpr auto forward_sweep = [&]() {
-            StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-            forward_sweep_(I0) = true;
-
-            // TODO: BUG: should start at 1
-            static_for<1, nDim, 1>{}([&](auto i) {
-                index_t tmp = ordered_src_access_lengths[I0] - 1;
-
-                static_for<1, i, 1>{}([&](auto j) {
-                    tmp = tmp * ordered_src_access_lengths[j] + ordered_src_access_lengths[j] - 1;
-                });
-
-                forward_sweep_(i) = tmp % 2 == 0;
-            });
-
-            return forward_sweep_;
-        }();
-
-        // calculate src data index after last iteration in RunRead(), if it has not being reset by
-        // RunRead()
-        constexpr auto src_data_idx = [&]() {
-            Index ordered_idx;
-
-            static_for<0, nDim, 1>{}([&](auto i) {
-                ordered_idx(i) = forward_sweep[i] ? ordered_src_access_lengths[i] - 1 : 0;
-            });
-
-            return container_reorder_given_old2new(ordered_idx, src_dim_access_order) *
-                   src_scalar_per_access;
-        }();
-
-        //
-        constexpr auto reset_src_data_step = [&]() {
-            Index reset_src_data_step_;
-
-            static_for<0, nDim, 1>{}([&](auto i) { reset_src_data_step_(i) = -src_data_idx[i]; });
-
-            return reset_src_data_step_;
-        }();
-
-        return reset_src_data_step;
-    }
-
-    __device__ static constexpr auto GetDstCoordinateResetStep()
-    {
-        constexpr auto I0 = Number<0>{};
-
-        // scalar per access on each dim
-        // TODO: don't use lambda_scalar_per_access
-        constexpr auto dst_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<DstVectorDim, DstScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto dst_access_lengths = SliceLengths{} / dst_scalar_per_access;
-
-        constexpr auto dst_dim_access_order = DstDimAccessOrder{};
-
-        constexpr auto ordered_dst_access_lengths =
-            container_reorder_given_new2old(dst_access_lengths, dst_dim_access_order);
-
-        // judge move forward or move backward during the last iteration
-        constexpr auto forward_sweep = [&]() {
-            StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-            forward_sweep_(I0) = true;
-
-            static_for<1, nDim, 1>{}([&](auto i) {
-                index_t tmp = ordered_dst_access_lengths[I0] - 1;
-
-                static_for<1, i, 1>{}([&](auto j) {
-                    tmp = tmp * ordered_dst_access_lengths[j] + ordered_dst_access_lengths[j] - 1;
-                });
-
-                forward_sweep_(i) = tmp % 2 == 0;
-            });
-
-            return forward_sweep_;
-        }();
-
-        // calculate dst data index after last iteration in RunWrite(), if it has not being reset by
-        // RunWrite()
-        constexpr auto dst_data_idx = [&]() {
-            Index ordered_idx;
-
-            static_for<0, nDim, 1>{}([&](auto i) {
-                ordered_idx(i) = forward_sweep[i] ? ordered_dst_access_lengths[i] - 1 : 0;
-            });
-
-            return container_reorder_given_old2new(ordered_idx, dst_dim_access_order) *
-                   dst_scalar_per_access;
-        }();
-
-        //
-        constexpr auto reset_dst_data_step = [&]() {
-            Index reset_dst_data_step_;
-
-            static_for<0, nDim, 1>{}([&](auto i) { reset_dst_data_step_(i) = -dst_data_idx[i]; });
-
-            return reset_dst_data_step_;
-        }();
-
-        return reset_dst_data_step;
-    }
-
-    // src_slice_origin_step_idx need to be known at compile-time, for performance reason
-    __device__ void MoveSrcSliceWindow(const SrcDesc& src_desc,
-                                       const Index& src_slice_origin_step_idx)
-    {
-        // if src coord was not reset by RunRead(), then need to adjust the step here
-        const auto adjusted_step_idx =
-            SrcResetCoordinateAfterRun ? src_slice_origin_step_idx
-                                       : src_slice_origin_step_idx + GetSrcCoordinateResetStep();
-
-        // is it OK to construct a new step every time?
-        const auto adjusted_step = make_tensor_coordinate_step(src_desc, adjusted_step_idx);
-
-        move_tensor_coordinate(src_desc, src_coord_, adjusted_step);
-    }
-
-    // src_slice_origin_step_idx need to be known at compile-time, for performance reason
-    __device__ void MoveSrcSliceWindow(const SrcDesc& src_desc,
-                                       const Index& src_slice_origin_step_idx)
-    {
-        // if src coord was not reset by RunRead(), then need to adjust the step here
-        const auto adjusted_step_idx =
-            SrcResetCoordinateAfterRun ? src_slice_origin_step_idx
-                                       : src_slice_origin_step_idx + GetSrcCoordinateResetStep();
-
-        // is it OK to construct a new step every time?
-        const auto adjusted_step = make_tensor_coordinate_step(src_desc, adjusted_step_idx);
-
-        move_tensor_coordinate(src_desc, src_coord_, adjusted_step);
-    }
-
-    // dst_slice_origin_step_idx need to be known at compile-time, for performance reason
-    __device__ void MoveDstSliceWindow(const DstDesc& dst_desc,
-                                       const Dst0Desc dst0_desc,
-                                       const Dst1Desc dst1_desc,
-                                       const Index& dst_slice_origin_step_idx)
-    {
-        // if dst coord was not reset by RunWrite(), then need to adjust the step here
-        const auto adjusted_step_idx =
-            DstResetCoordinateAfterRun ? dst_slice_origin_step_idx
-                                       : dst_slice_origin_step_idx + GetDstCoordinateResetStep();
-
-        // is it OK to construct a new step every time?
-        const auto adjusted_step = make_tensor_coordinate_step(dst_desc, adjusted_step_idx);
-
-        move_tensor_coordinate(dst_desc, dst_coord_, adjusted_step);
-        move_tensor_coordinate(dst0_desc, dst0_coord_, adjusted_step);
-        move_tensor_coordinate(dst1_desc, dst1_coord_, adjusted_step);
-    }
-
-    __device__ static constexpr auto GetSrcThreadScratchDescriptor()
-    {
-        constexpr auto src_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<SrcVectorDim, SrcScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto src_access_lengths = SliceLengths{} / src_scalar_per_access;
-
-        constexpr auto src_access_lengths_and_vector_length = container_push_back(
-            sequence_to_tuple_of_number(src_access_lengths), Number<SrcScalarPerVector>{});
-
-        // 1st stage of transforms
-        constexpr auto desc0 =
-            make_naive_tensor_descriptor_packed(src_access_lengths_and_vector_length);
-
-        // 2nd stage of transforms
-        constexpr auto transforms = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == SrcVectorDim)
-                {
-                    return make_merge_transform_v3_division_mod(
-                        make_tuple(src_access_lengths_and_vector_length[i],
-                                   src_access_lengths_and_vector_length[Number<nDim>{}]));
-                }
-                else
-                {
-                    return make_pass_through_transform(src_access_lengths_and_vector_length[i]);
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto low_dim_idss = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == SrcVectorDim)
-                {
-                    return Sequence<i.value, nDim>{};
-                }
-                else
-                {
-                    return Sequence<i.value>{};
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto up_dim_idss =
-            generate_tuple([&](auto i) { return Sequence<i.value>{}; }, Number<nDim>{});
-
-        return transform_tensor_descriptor(desc0, transforms, low_dim_idss, up_dim_idss);
-    }
-
-    __device__ static constexpr auto GetDstThreadScratchDescriptor()
-    {
-        // 1st stage of transforms
-        constexpr auto dst_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<DstVectorDim, DstScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto dst_access_lengths = SliceLengths{} / dst_scalar_per_access;
-
-        constexpr auto dst_access_lengths_and_vector_length = container_push_back(
-            sequence_to_tuple_of_number(dst_access_lengths), Number<DstScalarPerVector>{});
-
-        constexpr auto desc0 =
-            make_naive_tensor_descriptor_packed(dst_access_lengths_and_vector_length);
-
-        // 2nd stage of transforms
-        constexpr auto transforms = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == DstVectorDim)
-                {
-                    return make_merge_transform_v3_division_mod(
-                        make_tuple(dst_access_lengths_and_vector_length[i],
-                                   dst_access_lengths_and_vector_length[Number<nDim>{}]));
-                }
-                else
-                {
-                    return make_pass_through_transform(dst_access_lengths_and_vector_length[i]);
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto low_dim_idss = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == DstVectorDim)
-                {
-                    return Sequence<i.value, nDim>{};
-                }
-                else
-                {
-                    return Sequence<i.value>{};
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto up_dim_idss =
-            generate_tuple([&](auto i) { return Sequence<i.value>{}; }, Number<nDim>{});
-
-        return transform_tensor_descriptor(desc0, transforms, low_dim_idss, up_dim_idss);
-    }
-
-    private:
-    static constexpr auto src_thread_scratch_desc_ = decltype(GetSrcThreadScratchDescriptor()){};
-    static constexpr auto dst_thread_scratch_desc_ = decltype(GetDstThreadScratchDescriptor()){};
-
-    StaticTensorTupleOfVectorBuffer<AddressSpaceEnum::Vgpr,
-                                    SrcData,
-                                    SrcScalarPerVector,
-                                    decltype(src_thread_scratch_desc_),
-                                    true>
-        src_thread_scratch_;
-
-    StaticTensorTupleOfVectorBuffer<AddressSpaceEnum::Vgpr,
-                                    DstData,
-                                    DstScalarPerVector,
-                                    decltype(dst_thread_scratch_desc_),
-                                    true>
-        dst_thread_scratch_;
-
-    SrcCoord src_coord_;
-    DstCoord dst_coord_;
-    const SrcElementwiseOperation src_element_op_;
-    const DstElementwiseOperation dst_element_op_;
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp
index 6e8a23930bbb91677ee18bab216af6c45de72e4c..6a6c1f2ac5bfed7f02c5f5895d16002226cff56c 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp
index f13da341f9b2d36d992b1cd1835b43c89dcabd1f..bd01108b03ca82d4877e19806521efda8da8c6c3 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
index 9c91cd9ca8f86df37fee6dfdf597be1cefe2d683..6ec9abc41708735011acda7fcd01e54cbec5c4a6 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp
index 68bc2726f4be6fa1b08fe9213e0db9440526f975..cf2c7a2aee3d3f612fd12c0868477bfdfe49a2ea 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp
index 0f5fb88b04540dbf355fb0f02998303b7680c193..b5847e51b42ea5b567133f0307b319fa3d8fadbb 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp
index 2eb1b0ee90a446ee3b14354aff018275cf7809f4..db7dee21992d0e4edaa2e7a443b5c0ed68589107 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp
index 12ba2c5381311ab45ffad4be793651d176f5fc59..eb6715e8ebb382e438791f60193bfc23f06d1773 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp b/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..979f3567e987ed66fd45af5deeae7bb2ac3fc761
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
@@ -0,0 +1,507 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/utility/math.hpp"
+#include "ck/utility/amd_wmma.hpp"
+
+namespace ck {
+
+enum struct WmmaInstr
+{
+    wmma_f32_16x16x16_f16 = 0,
+    wmma_f32_16x16x16_bf16,
+    wmma_f16_16x16x16_f16,
+    wmma_bf16_16x16x16_bf16,
+    wmma_i32_16x16x16_iu8,
+    wmma_i32_16x16x16_iu4
+};
+
+/*
+ *  WMMA Wave Tile Always MxNxK = 16x16x16
+ *  WAVE32
+        -----------------------------------
+        |RC0| | | | | | | | | | | | | | | |	   SubGroup 0
+        |RC1| | | | | | | | | | | | | | | |
+        |RC2| | | | | | | | | | | | | | | |
+        |RC3|T|T|T|T|T|T|T|T|T|T|T|T|T|T|T|
+        |RC4|0|0|0|0|0|0|0|0|0|1|1|1|1|1|1|
+        |RC5|1|2|3|4|5|6|7|8|9|0|1|2|3|4|5|
+        |RC6| | | | | | | | | | | | | | | |
+        |RC7| | | | | | | | | | | | | | | |
+        -----------------------------------
+        |   | | | | | | | | | | | | | | | |	   SubGroup 1
+        |   | | | | | | | | | | | | | | | |
+        | T |T|T|T|T|T|T|T|T|T|T|T|T|T|T|T|
+        | 1 |1|1|1|2|2|2|2|2|2|2|2|2|2|3|3|
+        | 6 |7|8|9|0|1|2|3|4|5|6|7|8|9|0|1|
+        |   | | | | | | | | | | | | | | | |
+        |   | | | | | | | | | | | | | | | |
+        |   | | | | | | | | | | | | | | | |
+        -----------------------------------
+
+
+ *  WAVE64
+        -----------------------------------
+        |RC0|T|T|T|T|T|T|T|T|T|T|T|T|T|T|T|	   SubGroup 0
+        |RC1|0|0|0|0|0|0|0|0|0|1|1|1|1|1|1|
+        |RC2|1|2|3|4|5|6|7|8|9|0|1|2|3|4|5|
+        |RC3|T|T|T|T|T|T|T|T|T|T|T|T|T|T|T|
+        -----------------------------------
+        | T |T|T|T|T|T|T|T|T|T|T|T|T|T|T|T|    SubGroup 1
+        | 1 |1|1|1|2|2|2|2|2|2|2|2|2|2|3|3|
+        | 6 |7|8|9|0|1|2|3|4|5|6|7|8|9|0|1|
+        |   | | | | | | | | | | | | | | | |
+        -----------------------------------
+        | T |T|T|T|T|T|T|T|T|T|T|T|T|T|T|T|	   SubGroup 2
+        | 3 |3|3|3|3|3|3|3|4|4|4|4|4|4|4|4|
+        | 2 |3|4|5|6|7|8|9|0|1|2|3|4|5|6|7|
+        |   | | | | | | | | | | | | | | | |
+        -----------------------------------
+        | T |T|T|T|T|T|T|T|T|T|T|T|T|T|T|T|    SubGroup 3
+        | 4 |4|5|5|5|5|5|5|5|5|5|5|6|6|6|6|
+        | 8 |9|0|1|2|3|4|5|6|7|8|9|0|1|2|3|
+        |   | | | | | | | | | | | | | | | |
+        -----------------------------------
+
+*   RC = Register for storing accumalted result
+*	T  = Thread ID
+*/
+
+template <WmmaInstr Instr, index_t WaveSize, typename = void>
+struct wmma_type
+{
+};
+
+// A-swizzled
+template <index_t WaveSize>
+struct wmma_type<WmmaInstr::wmma_f32_16x16x16_f16,
+                 WaveSize,
+                 typename std::enable_if_t<WaveSize == 32 || WaveSize == 64>>
+{
+    // Absolute fixing property
+    // * Data Pixel
+    static constexpr index_t m_per_wmma      = 16;
+    static constexpr index_t n_per_wmma      = 16;
+    static constexpr index_t k_per_wmma      = 16;
+    static constexpr index_t src_a_data_size = 2;
+    static constexpr index_t src_b_data_size = 2;
+    static constexpr index_t acc_data_size   = 4;
+    // * Thread mapping inside wave, num_thread_per_subgroups always alone N direction
+    static constexpr index_t num_thread_per_subgroups = n_per_wmma;
+
+    // Wave mode dependent propety
+    static constexpr index_t wave_size = Number<WaveSize>{};
+    // * Fixed in Navi3x, Will be wave mode dependent on Navi4x
+    static constexpr index_t num_src_a_vgprs_per_wave = m_per_wmma * src_a_data_size / 4;
+    static constexpr index_t num_src_b_vgprs_per_wave = n_per_wmma * src_b_data_size / 4;
+    // * num_acc_vgprs_per_wave alone M direction
+    // * num_subgroups alone M direction
+    static constexpr index_t num_acc_vgprs_per_wave =
+        m_per_wmma * n_per_wmma * acc_data_size / wave_size / 4;
+    static constexpr index_t num_subgroups = wave_size / num_thread_per_subgroups;
+
+    template <index_t MPerWmma, index_t NPerWmma, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        if constexpr(wave_size == 32)
+        {
+            intrin_wmma_f32_16x16x16_f16_w32<MPerWmma, NPerWmma>::Run(a, b, reg_c);
+        }
+        else if constexpr(wave_size == 64)
+        {
+            intrin_wmma_f32_16x16x16_f16_w64<MPerWmma, NPerWmma>::Run(a, b, reg_c);
+        }
+    }
+};
+
+template <index_t WaveSize>
+struct wmma_type<WmmaInstr::wmma_f32_16x16x16_bf16,
+                 WaveSize,
+                 typename std::enable_if_t<WaveSize == 32 || WaveSize == 64>>
+{
+    // Absolute fixing property
+    static constexpr index_t m_per_wmma               = 16;
+    static constexpr index_t n_per_wmma               = 16;
+    static constexpr index_t k_per_wmma               = 16;
+    static constexpr index_t src_a_data_size          = 2;
+    static constexpr index_t src_b_data_size          = 2;
+    static constexpr index_t acc_data_size            = 4;
+    static constexpr index_t num_thread_per_subgroups = n_per_wmma;
+
+    // Wave mode dependent propety
+    static constexpr index_t wave_size                = Number<WaveSize>{};
+    static constexpr index_t num_src_a_vgprs_per_wave = m_per_wmma * src_a_data_size / 4;
+    static constexpr index_t num_src_b_vgprs_per_wave = n_per_wmma * src_b_data_size / 4;
+    static constexpr index_t num_acc_vgprs_per_wave =
+        m_per_wmma * n_per_wmma * acc_data_size / wave_size / 4;
+    static constexpr index_t num_subgroups = wave_size / num_thread_per_subgroups;
+
+    template <index_t MPerWmma, index_t NPerWmma, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        if constexpr(wave_size == 32)
+        {
+            intrin_wmma_f32_16x16x16_bf16_w32<MPerWmma, NPerWmma>::Run(a, b, reg_c);
+        }
+        else if constexpr(wave_size == 64)
+        {
+            intrin_wmma_f32_16x16x16_bf16_w64<MPerWmma, NPerWmma>::Run(a, b, reg_c);
+        }
+    }
+};
+
+#ifdef CK_UNPACKED_ACC_DESC_LOGIC
+template <index_t WaveSize>
+struct wmma_type<WmmaInstr::wmma_f16_16x16x16_f16,
+                 WaveSize,
+                 typename std::enable_if_t<WaveSize == 32 || WaveSize == 64>>
+{
+    // Absolute fixing property
+    static constexpr index_t m_per_wmma               = 16;
+    static constexpr index_t n_per_wmma               = 16;
+    static constexpr index_t k_per_wmma               = 16;
+    static constexpr index_t src_a_data_size          = 2;
+    static constexpr index_t src_b_data_size          = 2;
+    static constexpr index_t acc_data_size            = 2;
+    static constexpr index_t num_thread_per_subgroups = n_per_wmma;
+
+    // Wave mode dependent propety
+    static constexpr index_t wave_size                = Number<WaveSize>{};
+    static constexpr index_t num_src_a_vgprs_per_wave = m_per_wmma * src_a_data_size / 4;
+    static constexpr index_t num_src_b_vgprs_per_wave = n_per_wmma * src_b_data_size / 4;
+    static constexpr index_t num_acc_vgprs_per_wave =
+        m_per_wmma * n_per_wmma * acc_data_size / wave_size / 4;
+    static constexpr index_t num_subgroups = wave_size / num_thread_per_subgroups;
+
+    template <index_t MPerWmma,
+              index_t NPerWmma,
+              index_t Opsel,
+              class FloatA,
+              class FloatB,
+              class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        if constexpr(wave_size == 32)
+        {
+            intrin_wmma_f16_16x16x16_f16_w32<MPerWmma, NPerWmma, Opsel>::Run(a, b, reg_c);
+        }
+        else if constexpr(wave_size == 64)
+        {
+            intrin_wmma_f16_16x16x16_f16_w64<MPerWmma, NPerWmma, Opsel>::Run(a, b, reg_c);
+        }
+    }
+};
+
+template <index_t WaveSize>
+struct wmma_type<WmmaInstr::wmma_bf16_16x16x16_bf16,
+                 WaveSize,
+                 typename std::enable_if_t<WaveSize == 32 || WaveSize == 64>>
+{
+    // Absolute fixing property
+    static constexpr index_t m_per_wmma               = 16;
+    static constexpr index_t n_per_wmma               = 16;
+    static constexpr index_t k_per_wmma               = 16;
+    static constexpr index_t src_a_data_size          = 2;
+    static constexpr index_t src_b_data_size          = 2;
+    static constexpr index_t acc_data_size            = 2;
+    static constexpr index_t num_thread_per_subgroups = n_per_wmma;
+
+    // Wave mode dependent propety
+    static constexpr index_t wave_size                = Number<WaveSize>{};
+    static constexpr index_t num_src_a_vgprs_per_wave = m_per_wmma * src_a_data_size / 4;
+    static constexpr index_t num_src_b_vgprs_per_wave = n_per_wmma * src_b_data_size / 4;
+    static constexpr index_t num_acc_vgprs_per_wave =
+        m_per_wmma * n_per_wmma * acc_data_size / wave_size / 4;
+    static constexpr index_t num_subgroups = wave_size / num_thread_per_subgroups;
+
+    template <index_t MPerWmma,
+              index_t NPerWmma,
+              index_t Opsel,
+              class FloatA,
+              class FloatB,
+              class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        if constexpr(wave_size == 32)
+        {
+            intrin_wmma_bf16_16x16x16_bf16_w32<MPerWmma, NPerWmma, Opsel>::Run(a, b, reg_c);
+        }
+        else if constexpr(wave_size == 64)
+        {
+            intrin_wmma_bf16_16x16x16_bf16_w64<MPerWmma, NPerWmma, Opsel>::Run(a, b, reg_c);
+        }
+    }
+};
+
+#endif
+
+template <index_t WaveSize>
+struct wmma_type<WmmaInstr::wmma_i32_16x16x16_iu8,
+                 WaveSize,
+                 typename std::enable_if_t<WaveSize == 32 || WaveSize == 64>>
+{
+    // Absolute fixing property
+    static constexpr index_t m_per_wmma               = 16;
+    static constexpr index_t n_per_wmma               = 16;
+    static constexpr index_t k_per_wmma               = 16;
+    static constexpr index_t src_a_data_size          = 2;
+    static constexpr index_t src_b_data_size          = 2;
+    static constexpr index_t acc_data_size            = 4;
+    static constexpr index_t num_thread_per_subgroups = n_per_wmma;
+
+    // Wave mode dependent propety
+    static constexpr index_t wave_size                = Number<WaveSize>{};
+    static constexpr index_t num_src_a_vgprs_per_wave = m_per_wmma * src_a_data_size / 4;
+    static constexpr index_t num_src_b_vgprs_per_wave = n_per_wmma * src_b_data_size / 4;
+    static constexpr index_t num_acc_vgprs_per_wave =
+        m_per_wmma * n_per_wmma * acc_data_size / wave_size / 4;
+    static constexpr index_t num_subgroups = wave_size / num_thread_per_subgroups;
+
+    template <index_t MPerWmma,
+              index_t NPerWmma,
+              class FloatA,
+              class FloatB,
+              class FloatC,
+              bool neg_a = false,
+              bool neg_b = false,
+              bool clamp = false>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        if constexpr(wave_size == 32)
+        {
+            intrin_wmma_i32_16x16x16_iu8_w32<MPerWmma, NPerWmma, neg_a, neg_b, clamp>::Run(
+                a, b, reg_c);
+        }
+        else if constexpr(wave_size == 64)
+        {
+            intrin_wmma_i32_16x16x16_iu8_w64<MPerWmma, NPerWmma, neg_a, neg_b, clamp>::Run(
+                a, b, reg_c);
+        }
+    }
+};
+
+template <typename src_type_a,
+          typename src_type_b,
+          typename dst_type,
+          index_t MPerWmma,
+          index_t NPerWmma>
+struct WmmaSelector
+{
+    template <typename src_type_a_,
+              typename src_type_b_,
+              typename dst_type_,
+              index_t MPerWmma_,
+              index_t NPerWmma_>
+    static constexpr auto GetWmma();
+
+    template <>
+    static constexpr auto GetWmma<half_t, half_t, float, 16, 16>()
+    {
+        return WmmaInstr::wmma_f32_16x16x16_f16;
+    }
+
+    template <>
+    static constexpr auto GetWmma<bhalf_t, bhalf_t, float, 16, 16>()
+    {
+        return WmmaInstr::wmma_f32_16x16x16_bf16;
+    }
+
+    template <>
+    static constexpr auto GetWmma<half_t, half_t, half_t, 16, 16>()
+    {
+        return WmmaInstr::wmma_f16_16x16x16_f16;
+    }
+
+    template <>
+    static constexpr auto GetWmma<bhalf_t, bhalf_t, bhalf_t, 16, 16>()
+    {
+        return WmmaInstr::wmma_bf16_16x16x16_bf16;
+    }
+
+    template <>
+    static constexpr auto GetWmma<int8_t, int8_t, int, 16, 16>()
+    {
+        return WmmaInstr::wmma_i32_16x16x16_iu8;
+    }
+#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
+    template <>
+    static constexpr auto GetWmma<int4_t, int, 16, 16>()
+    {
+        return WmmaInstr::wmma_i32_16x16x16_iu4;
+    }
+#endif
+    // get_warp_size do not return the correct wavesize, hardcode to 32 as workaround
+    static constexpr auto selected_wmma =
+        wmma_type<GetWmma<src_type_a, src_type_b, dst_type, MPerWmma, NPerWmma>(), Number<32>{}>{};
+
+    __host__ __device__ constexpr WmmaSelector()
+    {
+        static_assert(selected_wmma.m_per_wmma == 16, "WRONG! WMMA_M must equal to 16");
+
+        static_assert(selected_wmma.m_per_wmma == 16, "WRONG! WMMA_M must equal to 16");
+
+        static_assert(selected_wmma.k_per_wmma == 16, "WRONG! WMMA_M must equal to 16");
+
+        static_assert(selected_wmma.wave_size * selected_wmma.num_acc_vgprs_per_wave *
+                              selected_wmma.acc_data_size ==
+                          selected_wmma.m_per_wmma * selected_wmma.n_per_wmma * 4,
+                      "WRONG! Invalid Number of Accumulator Register");
+    }
+};
+
+template <typename src_type_a,
+          typename src_type_b,
+          typename dst_type,
+          index_t MPerWmma,
+          index_t NPerWmma,
+          index_t KPack,
+          bool TransposeC = false>
+struct WmmaGemm
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    using CIndex   = MultiIndex<2>;
+    using CIndex4D = MultiIndex<4>;
+
+    __host__ __device__ constexpr WmmaGemm()
+    {
+        static_assert(NPerWmma == 16 && MPerWmma == 16,
+                      "Only support GemmNPerWmma == 16 and GemmMPerWmma == 16 for wmma");
+
+        static_assert(KPack == wmma_instr.k_per_wmma, "KPack should be k_per_wmma");
+    }
+
+    // WMMA output supporting C = A * B
+    // Vector Write
+    // MPerWMMA_NPerWMMA -> MSubGroup_..._NPerWMMA_MAccVgprPerWave
+    template <typename CDesc_MBlockxRepeat_MWave_MPerWMMA_NBlockxRepeat_NWave_NPerWMMA>
+    __host__ __device__ static constexpr auto
+    MakeCDesc_MBlockxRepeat_MWave_MSubGroup_NBlockxRepeat_NWave_NThreadPerSubGroup_MAccVgprs(
+        const CDesc_MBlockxRepeat_MWave_MPerWMMA_NBlockxRepeat_NWave_NPerWMMA&
+            c_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma)
+    {
+        const auto MBlockxRepeat =
+            c_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma.GetLength(I0);
+        const auto NBlockxRepeat =
+            c_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma.GetLength(I3);
+        const auto MWave =
+            c_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma.GetLength(I1);
+        const auto NWave =
+            c_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma.GetLength(I4);
+
+        return transform_tensor_descriptor(
+            c_desc_mblockxrepeat_mwave_mperwmma_nblockxrepeat_nwave_nperwmma,
+            make_tuple(
+                make_pass_through_transform(MBlockxRepeat),
+                make_pass_through_transform(MWave),
+                make_unmerge_transform(make_tuple(Number<wmma_instr.num_subgroups>{},
+                                                  Number<wmma_instr.num_acc_vgprs_per_wave>{})),
+                make_pass_through_transform(NBlockxRepeat),
+                make_pass_through_transform(NWave),
+                make_pass_through_transform(Number<wmma_instr.num_thread_per_subgroups>{})),
+            make_tuple(Sequence<0>{},
+                       Sequence<1>{},
+                       Sequence<2>{},
+                       Sequence<3>{},
+                       Sequence<4>{},
+                       Sequence<5>{}),
+            make_tuple(Sequence<0>{},
+                       Sequence<1>{},
+                       Sequence<2, 6>{},
+                       Sequence<3>{},
+                       Sequence<4>{},
+                       Sequence<5>{}));
+    }
+
+    __device__ static constexpr index_t GetRegSizePerWmma()
+    {
+        return wmma_instr.num_acc_vgprs_per_wave;
+    }
+
+    __device__ static constexpr index_t GetWaveSize() { return wmma_instr.wave_size; }
+
+    template <class FloatA, class FloatB, class FloatC>
+    __device__ void Run(const FloatA& p_a_wave, const FloatB& p_b_wave, FloatC& p_c_thread) const
+    {
+        static_assert(
+            (is_same<src_type_a, half_t>::value && is_same<src_type_b, half_t>::value &&
+             is_same<dst_type, float>::value) ||
+                (is_same<src_type_a, bhalf_t>::value && is_same<src_type_b, bhalf_t>::value &&
+                 is_same<dst_type, float>::value) ||
+                (is_same<src_type_a, half_t>::value && is_same<src_type_b, half_t>::value &&
+                 is_same<dst_type, half_t>::value) ||
+                (is_same<src_type_a, bhalf_t>::value && is_same<src_type_b, bhalf_t>::value &&
+                 is_same<dst_type, bhalf_t>::value) ||
+                (is_same<src_type_a, int8_t>::value && is_same<src_type_b, int8_t>::value &&
+                 is_same<dst_type, int32_t>::value)
+#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
+                || (is_same<src_type_a, int4_t>::value && is_same<src_type_b, int4_t>::value &&
+                    is_same<dst_type, int32_t>::value)
+#endif
+                ,
+            "base type couple must be (half, float), (bhalf, float), (half, half), (bhalf, bhalf), "
+            "(int8, int32) or (int4, int32)!");
+        if constexpr(!TransposeC)
+        {
+            wmma_instr.template run<MPerWmma, NPerWmma>(p_a_wave, p_b_wave, p_c_thread);
+        }
+        else
+        {
+            wmma_instr.template run<MPerWmma, NPerWmma>(p_b_wave, p_a_wave, p_c_thread);
+        }
+    }
+
+    __device__ static auto GetLaneId() { return get_thread_local_1d_id() % wmma_instr.wave_size; }
+
+    __device__ static auto GetSubGroupId()
+    {
+        return (GetLaneId() / wmma_instr.num_thread_per_subgroups) % wmma_instr.num_subgroups;
+    }
+
+    __device__ static auto GetLaneIdUnderSubGroup()
+    {
+        return GetLaneId() % wmma_instr.num_thread_per_subgroups;
+    }
+    __device__ static auto GetSwizzledLaneIdLow()
+    {
+        return ((GetLaneIdUnderSubGroup() & 1) << 3) | (GetLaneIdUnderSubGroup() >> 1);
+    }
+
+    __host__ __device__ static auto CalculateAThreadOriginDataIndex()
+    {
+        return GetSwizzledLaneIdLow();
+    }
+
+    __host__ __device__ static auto CalculateBThreadOriginDataIndex()
+    {
+        return GetLaneIdUnderSubGroup();
+    }
+
+    __device__ static CIndex GetBeginOfThreadBlk()
+    {
+        index_t n_offset = GetLaneIdUnderSubGroup();
+        index_t m_offset = GetSubGroupId() * wmma_instr.num_acc_vgprs_per_wave;
+
+        return TransposeC ? CIndex{n_offset, m_offset} : CIndex{m_offset, n_offset};
+    }
+
+    static constexpr auto wmma =
+        WmmaSelector<src_type_a, src_type_b, dst_type, MPerWmma, NPerWmma>{};
+    static constexpr auto wmma_instr = wmma.selected_wmma;
+
+    __host__ __device__ static constexpr auto
+    GetCMSubGroupNThreadPerSubGroupMAccVgprsThreadBlkLengths()
+    {
+        return make_tuple(I1, I1, Number<wmma_instr.num_acc_vgprs_per_wave>{});
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index 4d53f0d81620141ff6dfc6cb965209709341aaf8..814969ef42baa2cc8e1f4188f61b3d50d74ea265 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -27,7 +27,11 @@ enum struct MfmaInstr
     mfma_f32_16x16x8bf16,
     mfma_i32_32x32x8i8,
     mfma_i32_16x16x16i8,
-    mfma_f64_16x16x4f64
+    mfma_i32_32x32x16i8,
+    mfma_i32_16x16x32i8,
+    mfma_f64_16x16x4f64,
+    mfma_f32_32x32x16f8f8,
+    mfma_f32_16x16x32f8f8
 };
 
 template <MfmaInstr instr>
@@ -386,6 +390,50 @@ struct mfma_type<MfmaInstr::mfma_i32_16x16x16i8>
     }
 };
 
+template <>
+struct mfma_type<MfmaInstr::mfma_i32_32x32x16i8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 4;
+    static constexpr index_t num_regs_per_blk    = 16;
+    static constexpr index_t num_threads_per_blk = 32;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 2;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 32;
+    static constexpr index_t n_per_blk           = 32;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_i32_32x32x16i8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct mfma_type<MfmaInstr::mfma_i32_16x16x32i8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 1;
+    static constexpr index_t num_regs_per_blk    = 4;
+    static constexpr index_t num_threads_per_blk = 16;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 4;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 16;
+    static constexpr index_t n_per_blk           = 16;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_i32_16x16x32i8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
 template <>
 struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64>
 {
@@ -408,6 +456,50 @@ struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64>
     }
 };
 
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_32x32x16f8f8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 4;
+    static constexpr index_t num_regs_per_blk    = 16;
+    static constexpr index_t num_threads_per_blk = 32;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 2;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 32;
+    static constexpr index_t n_per_blk           = 32;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_32x32x16f8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8f8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 1;
+    static constexpr index_t num_regs_per_blk    = 4;
+    static constexpr index_t num_threads_per_blk = 16;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 4;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 16;
+    static constexpr index_t n_per_blk           = 16;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_16x16x32f8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
 template <typename base_type, index_t MPerXdlops, index_t NPerXdlops>
 struct MfmaSelector
 {
@@ -524,17 +616,41 @@ struct MfmaSelector
 #endif
     }
 
+#if defined(CK_USE_AMD_MFMA_GFX940)
+    template <>
+    static constexpr auto GetMfma<int8_t, 32, 32>()
+    {
+        return MfmaInstr::mfma_i32_32x32x16i8;
+    }
+    template <>
+    static constexpr auto GetMfma<int8_t, 16, 16>()
+    {
+        return MfmaInstr::mfma_i32_16x16x32i8;
+    }
+#else
     template <>
     static constexpr auto GetMfma<int8_t, 32, 32>()
     {
         return MfmaInstr::mfma_i32_32x32x8i8;
     }
-
     template <>
     static constexpr auto GetMfma<int8_t, 16, 16>()
     {
         return MfmaInstr::mfma_i32_16x16x16i8;
     }
+#endif
+
+    template <>
+    static constexpr auto GetMfma<f8_t, 32, 32>()
+    {
+        return MfmaInstr::mfma_f32_32x32x16f8f8;
+    }
+
+    template <>
+    static constexpr auto GetMfma<f8_t, 16, 16>()
+    {
+        return MfmaInstr::mfma_f32_16x16x32f8f8;
+    }
 
     static constexpr auto selected_mfma = mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops>()>{};
 
@@ -736,7 +852,7 @@ struct XdlopsGemm
     {
         static_assert(is_same<base_type, double>::value || is_same<base_type, float>::value ||
                           is_same<base_type, half_t>::value || is_same<base_type, bhalf_t>::value ||
-                          is_same<base_type, int8_t>::value,
+                          is_same<base_type, int8_t>::value || is_same<base_type, f8_t>::value,
                       "base base_type must be double, float, half, bfloat16, and int8_t!");
 
         static_for<0, KPack / mfma_instr.k_per_blk, 1>{}([&](auto k) {
diff --git a/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp b/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp
index 5fc11d9158ab9a1261c1285c2cd9f5208d7afbf3..ea27a40ce3c7a48564af112a97af342c157f746d 100644
--- a/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
index 13d0a28cfe5b2db14a83a4b81e0bb379aabbe137..59f64bde77a0b2f3fb1b022a8951b3fecb91442f 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -13,6 +13,61 @@
 namespace ck {
 namespace tensor_operation {
 
+namespace {
+template <
+    index_t NDimSpatial,
+    typename ALayout,
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization ConvBwdDataSpecialization>
+constexpr auto
+make_out_n_ho_wo_k_grid_desc(const index_t N,
+                             const index_t Ho,
+                             const index_t Wo,
+                             const index_t K,
+                             const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_strides)
+{
+
+    if constexpr(is_same_v<ALayout, tensor_layout::convolution::NHWGK>)
+    {
+        const index_t NStride  = out_g_n_k_wos_strides[1];
+        const index_t HiStride = out_g_n_k_wos_strides[3];
+        const index_t WiStride = out_g_n_k_wos_strides[4];
+        const auto CStride     = Number<1>{};
+        if constexpr(ConvBwdDataSpecialization ==
+                     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
+                         Filter1x1Stride1Pad0)
+        {
+
+            return make_naive_tensor_descriptor(make_tuple(N * Ho * Wo, K),
+                                                make_tuple(WiStride, CStride));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor(make_tuple(N, Ho, Wo, K),
+                                                make_tuple(NStride, HiStride, WiStride, CStride));
+        }
+    }
+    else if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK>)
+    {
+        // assume packed
+        if constexpr(ConvBwdDataSpecialization ==
+                     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
+                         Filter1x1Stride1Pad0)
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N, Ho, Wo, K));
+        }
+    }
+    else
+    {
+        throw std::runtime_error("wrong! unsupported layout: " + ALayout::name());
+    }
+}
+
+} // namespace
+
 template <
     index_t NDimSpatial,
     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization ConvBwdDataSpecialization,
@@ -29,11 +84,12 @@ struct TransformConvBwdDataToGemm_v1
 
     template <typename ALayout,
               typename std::enable_if<NDimSpatial == 2 &&
-                                          is_same_v<ALayout, tensor_layout::convolution::GNHWK>,
+                                          (is_same_v<ALayout, tensor_layout::convolution::GNHWK> ||
+                                           is_same_v<ALayout, tensor_layout::convolution::NHWGK>),
                                       bool>::type = false>
     static auto MakeADescriptor_AK0_M_AK1(
         const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_lengths,
-        const std::array<index_t, NDimSpatial + 3>& /* out_g_n_k_wos_strides */,
+        const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_strides,
         const std::array<index_t, NDimSpatial + 3>& wei_g_k_c_xs_lengths,
         const std::array<index_t, NDimSpatial + 3>& /* wei_g_k_c_xs_strides */,
         const std::array<index_t, NDimSpatial + 3>& in_g_n_c_wis_lengths,
@@ -70,9 +126,9 @@ struct TransformConvBwdDataToGemm_v1
 
         const index_t AK0 = K / AK1;
 
-        // assume packed
         const auto out_n_ho_wo_k_grid_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(N, Ho, Wo, K));
+            make_out_n_ho_wo_k_grid_desc<NDimSpatial, ALayout, ConvBwdDataSpecialization>(
+                N, Ho, Wo, K, out_g_n_k_wos_strides);
 
         if constexpr(ConvBwdDataSpecialization ==
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
@@ -80,7 +136,7 @@ struct TransformConvBwdDataToGemm_v1
         {
             // A: output tensor
             const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
-                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
+                out_n_ho_wo_k_grid_desc,
                 make_tuple(make_pass_through_transform(N * Ho * Wo),
                            make_unmerge_transform(make_tuple(AK0, AK1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
index 1b5e64b66cf292fbcf4979d95aa4a5b1fa3e3a8b..cee3d2825b13facc912cdc8de21774ce793e9a2c 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
@@ -1,6 +1,6 @@
 
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/amd_address_space.hpp b/include/ck/utility/amd_address_space.hpp
index 9f1525914cdee16ee6cbed51297f0008d46f6085..d54f70e750e8032f706ae3f7f0dfec9f7cab4c8d 100644
--- a/include/ck/utility/amd_address_space.hpp
+++ b/include/ck/utility/amd_address_space.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index bdfb4f27580e011f0bebabfeae9146374a5ecd67..ea231154ea39a0db637049b6fb5a070c2ad5fb30 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "data_type.hpp"
@@ -286,7 +286,22 @@ llvm_amdgcn_raw_buffer_atomic_max_fp64(double vdata,
                                        int soffset,    // dst_wave_addr_offset
                                        int glc_slc) __asm("llvm.amdgcn.raw.buffer.atomic.fmax.f64");
 
-template <typename T, index_t N>
+// memory coherency bit for buffer store/load instruction
+// check ISA manual for each GFX target
+// e.g. for
+// https://www.amd.com/system/files/TechDocs/instinct-mi200-cdna2-instruction-set-architecture.pdf,
+// page 67~68
+enum struct AmdBufferCoherenceEnum
+{
+    DefaultCoherence = 0, // default value
+    GLC              = 1,
+    SLC              = 2,
+    GLC_SLC          = 3,
+};
+
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_wave_buffer_resource,
                                                                  index_t src_thread_addr_offset,
                                                                  index_t src_wave_addr_offset)
@@ -305,28 +320,37 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
         // use fp32 load to mimic fp64 load
         if constexpr(N == 1)
         {
-            const float2_t tmp = llvm_amdgcn_raw_buffer_load_fp32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            const float2_t tmp =
+                llvm_amdgcn_raw_buffer_load_fp32x2(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             return bit_cast<double>(tmp);
         }
         else if constexpr(N == 2)
         {
-            const float4_t tmp = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            const float4_t tmp =
+                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             return bit_cast<double2_t>(tmp);
         }
         else if constexpr(N == 4)
         {
-            const float4_t f32_0 = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            const float4_t f32_0 =
+                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             const float4_t f32_1 =
                 llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
                                                    src_thread_addr_offset,
                                                    src_wave_addr_offset + 4 * sizeof(float),
-                                                   0);
+                                                   static_cast<index_t>(coherence));
             vector_type<double, 4> tmp;
 
             tmp.AsType<double2_t>()(Number<0>{}) = bit_cast<double2_t>(f32_0);
@@ -339,31 +363,40 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_fp32(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp32(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_fp32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp32x2(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
             vector_type<float, 8> tmp;
 
-            tmp.AsType<float4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<float4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             tmp.AsType<float4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
                                                    src_thread_addr_offset,
                                                    src_wave_addr_offset + 4 * sizeof(float),
-                                                   0);
+                                                   static_cast<index_t>(coherence));
 
             return tmp.AsType<float8_t>()(Number<0>{});
         }
@@ -372,24 +405,32 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_fp16(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp16(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_fp16x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp16x2(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_fp16x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp16x4(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
             // use fp32 load to mimic fp16 load
-            float4_t tmp = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            float4_t tmp = llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<half8_t>(tmp);
         }
@@ -398,23 +439,31 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_i16(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i16(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_i16x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i16x2(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_i16x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i16x4(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
-            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<bhalf8_t>(tmp);
         }
@@ -423,31 +472,40 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_i32(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i32(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_i32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i32x2(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
             vector_type<int32_t, 8> tmp;
 
-            tmp.AsType<int32x4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<int32x4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                  src_thread_addr_offset,
+                                                  src_wave_addr_offset,
+                                                  static_cast<index_t>(coherence));
 
             tmp.AsType<int32x4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
                                                   src_thread_addr_offset,
                                                   src_wave_addr_offset + 4 * sizeof(int32_t),
-                                                  0);
+                                                  static_cast<index_t>(coherence));
             return tmp.AsType<int32x8_t>()(Number<0>{});
         }
     }
@@ -455,17 +513,23 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_i8(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i8(src_wave_buffer_resource,
+                                                  src_thread_addr_offset,
+                                                  src_wave_addr_offset,
+                                                  static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            return llvm_amdgcn_raw_buffer_load_i8x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i8x2(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
 #else
-            int16_t tmp = llvm_amdgcn_raw_buffer_load_i16(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int16_t tmp = llvm_amdgcn_raw_buffer_load_i16(src_wave_buffer_resource,
+                                                          src_thread_addr_offset,
+                                                          src_wave_addr_offset,
+                                                          static_cast<index_t>(coherence));
 
             return bit_cast<int8x2_t>(tmp);
 #endif
@@ -473,11 +537,15 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
         else if constexpr(N == 4)
         {
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            return llvm_amdgcn_raw_buffer_load_i8x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
 #else
-            int32_t tmp = llvm_amdgcn_raw_buffer_load_i32(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32_t tmp = llvm_amdgcn_raw_buffer_load_i32(src_wave_buffer_resource,
+                                                          src_thread_addr_offset,
+                                                          src_wave_addr_offset,
+                                                          static_cast<index_t>(coherence));
 
             return bit_cast<int8x4_t>(tmp);
 #endif
@@ -487,19 +555,24 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
             vector_type<int8_t, 8> tmp;
 
-            tmp.AsType<int8x4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_i8x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<int8x4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
+                                                 src_thread_addr_offset,
+                                                 src_wave_addr_offset,
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 4 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             return tmp.AsType<int8x8_t>()(Number<0>{});
 #else
-            int32x2_t tmp = llvm_amdgcn_raw_buffer_load_i32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32x2_t tmp = llvm_amdgcn_raw_buffer_load_i32x2(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<int8x8_t>(tmp);
 #endif
@@ -509,31 +582,36 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
             vector_type<int8_t, 16> tmp;
 
-            tmp.AsType<int8x4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_i8x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<int8x4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
+                                                 src_thread_addr_offset,
+                                                 src_wave_addr_offset,
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 4 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<2>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 8 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<3>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 12 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             return tmp.AsType<int8x16_t>()(Number<0>{});
 #else
-            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<int8x16_t>(tmp);
 #endif
@@ -541,7 +619,9 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     }
 }
 
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src_thread_data,
                                       int32x4_t dst_wave_buffer_resource,
                                       index_t dst_thread_addr_offset,
@@ -565,7 +645,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -573,7 +653,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, float>::value)
@@ -584,7 +664,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -592,7 +672,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -600,7 +680,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, half_t>::value)
@@ -611,7 +691,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -619,7 +699,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -627,7 +707,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
@@ -638,19 +718,19 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
 
             llvm_amdgcn_raw_buffer_store_fp16x4(tmp.AsType<half4_t>()[Number<1>{}],
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset + 4 * sizeof(half_t),
-                                                0);
+                                                static_cast<index_t>(coherence));
 #else
             llvm_amdgcn_raw_buffer_store_fp32x4(bit_cast<float4_t>(src_thread_data),
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
 #endif
         }
     }
@@ -662,7 +742,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -670,7 +750,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -678,7 +758,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
@@ -688,13 +768,13 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
 
             llvm_amdgcn_raw_buffer_store_i16x4(tmp.AsType<bhalf4_t>()[Number<1>{}],
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset + 4 * sizeof(bhalf_t),
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, int32_t>::value)
@@ -705,7 +785,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -713,7 +793,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -721,7 +801,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, int8_t>::value)
@@ -732,7 +812,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                             dst_wave_buffer_resource,
                                             dst_thread_addr_offset,
                                             dst_wave_addr_offset,
-                                            0);
+                                            static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -741,13 +821,13 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
 #else
             llvm_amdgcn_raw_buffer_store_i16(bit_cast<int16_t>(src_thread_data),
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
 #endif
         }
         else if constexpr(N == 4)
@@ -757,13 +837,13 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
 #else
             llvm_amdgcn_raw_buffer_store_i32(bit_cast<int32_t>(src_thread_data),
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
 #endif
         }
         else if constexpr(N == 8)
@@ -772,7 +852,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 16)
         {
@@ -780,7 +860,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
     }
 }
@@ -1012,7 +1092,9 @@ __device__ void amd_buffer_atomic_max_impl(const typename vector_type<T, N>::typ
 //   1) p_src_wave must point to global memory space
 //   2) p_src_wave must be a wavewise pointer.
 // It is user's responsibility to make sure that is true.
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ typename vector_type_maker<T, N>::type::type
 amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
                                             index_t src_thread_element_offset,
@@ -1032,13 +1114,30 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 #if CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK
     uint32_t src_addr_shift = src_thread_element_valid ? 0 : 0x80000000;
 
-    return amd_buffer_load_impl<scalar_t, vector_size>(
-        src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+    if constexpr(is_same<scalar_t, f8_t>::value)
+    {
+        auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+        return bit_cast<vector_t>(tmp);
+    }
+    else
+    {
+        return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+    }
 #else
-    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size>(
-        src_wave_buffer_resource, src_thread_addr_offset, 0);
-
-    return src_thread_element_valid ? tmp : vector_t(0);
+    if constexpr(is_same<scalar_t, f8_t>::value)
+    {
+        auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_thread_addr_offset, 0);
+        return src_thread_element_valid ? bit_cast<vector_t>(tmp) : vector_t(0);
+    }
+    else
+    {
+        vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_thread_addr_offset, 0);
+        return src_thread_element_valid ? tmp : vector_t(0);
+    }
 #endif
 }
 
@@ -1046,7 +1145,9 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 //   1) p_src_wave must point to global memory space
 //   2) p_src_wave must be a wavewise pointer.
 // It is user's responsibility to make sure that is true.
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ typename vector_type_maker<T, N>::type::type
 amd_buffer_load_invalid_element_return_customized_value(const T* p_src_wave,
                                                         index_t src_thread_element_offset,
@@ -1064,7 +1165,7 @@ amd_buffer_load_invalid_element_return_customized_value(const T* p_src_wave,
 
     constexpr index_t vector_size = scalar_type<vector_t>::vector_size;
 
-    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size>(
+    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
         src_wave_buffer_resource, src_thread_addr_offset, 0);
 
     return src_thread_element_valid ? tmp : vector_t(customized_value);
@@ -1074,7 +1175,9 @@ amd_buffer_load_invalid_element_return_customized_value(const T* p_src_wave,
 //   1) p_dst_wave must point to global memory
 //   2) p_dst_wave must be a wavewise pointer.
 // It is user's responsibility to make sure that is true.
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::type src_thread_data,
                                  T* p_dst_wave,
                                  const index_t dst_thread_element_offset,
@@ -1093,13 +1196,33 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
 #if CK_EXPERIMENTAL_USE_BUFFER_STORE_OOB_CHECK_OFFSET_TRICK
     uint32_t dst_addr_shift = dst_thread_element_valid ? 0 : 0x80000000;
 
-    amd_buffer_store_impl<scalar_t, vector_size>(
-        src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+    if constexpr(is_same<scalar_t, f8_t>::value)
+    {
+        auto tmp =
+            bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(src_thread_data);
+        amd_buffer_store_impl<int8_t, vector_size, coherence>(
+            tmp, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+    }
+    else
+    {
+        amd_buffer_store_impl<scalar_t, vector_size, coherence>(
+            src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+    }
 #else
     if(dst_thread_element_valid)
     {
-        amd_buffer_store_impl<scalar_t, vector_size>(
-            src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+        if constexpr(is_same<scalar_t, f8_t>::value)
+        {
+            auto tmp = bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
+                src_thread_data);
+            amd_buffer_store_impl<int8_t, vector_size, coherence>(
+                tmp, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+        }
+        else
+        {
+            amd_buffer_store_impl<scalar_t, vector_size, coherence>(
+                src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+        }
     }
 #endif
 }
diff --git a/include/ck/utility/amd_inline_asm.hpp b/include/ck/utility/amd_inline_asm.hpp
index 82bf2a5eb576c838f6f3d3c54b4b74c3838896d4..43baa817d3672c0b2eed6059a88351aeb19a5209 100644
--- a/include/ck/utility/amd_inline_asm.hpp
+++ b/include/ck/utility/amd_inline_asm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_AMD_INLINE_ASM_HPP
 #define CK_AMD_INLINE_ASM_HPP
@@ -220,8 +220,8 @@ amd_assembly_outer_product_1x2(int8x4_t a, int8x4_t b0, int8x4_t b1, int32_t& c0
                    "0"(c0),
                    "1"(c1));
 #else
-    c0 = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b0), c0, false);
-    c1 = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b1), c1, false);
+    c0     = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b0), c0, false);
+    c1     = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b1), c1, false);
 #endif
 }
 
@@ -257,10 +257,10 @@ __device__ void amd_assembly_outer_product_1x4(int8x4_t a,
                    "2"(c2),
                    "3"(c3));
 #else
-    c0 = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b0), c0, false);
-    c1 = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b1), c1, false);
-    c2 = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b2), c2, false);
-    c3 = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b3), c3, false);
+    c0     = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b0), c0, false);
+    c1     = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b1), c1, false);
+    c2     = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b2), c2, false);
+    c3     = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b3), c3, false);
 #endif
 }
 
@@ -355,5 +355,17 @@ __device__ void amd_assembly_outer_product_1x4(int8x16_t a,
                                    c3);
 }
 
+// Ranged input operand
+__device__ void amd_assembly_wmma_f32_16x16x16_f16_w32(half16_t a, half16_t b, float8_t& c)
+{
+#if defined(__gfx11__)
+    asm volatile("v_wmma_f32_16x16x16_f16 %0, %1, %2, %0" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+#else
+    ignore = a;
+    ignore = b;
+    ignore = c;
+#endif
+}
+
 } // namespace ck
 #endif
diff --git a/include/ck/utility/amd_llvm_intrinsic.hpp b/include/ck/utility/amd_llvm_intrinsic.hpp
deleted file mode 100644
index 01e77d7be897085b8c5816fe9e20fe924b859ba7..0000000000000000000000000000000000000000
--- a/include/ck/utility/amd_llvm_intrinsic.hpp
+++ /dev/null
@@ -1,14 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_AMD_LLVM_INTRINSIC_HPP
-#define CK_AMD_LLVM_INTRINSIC_HPP
-
-#include "data_type.hpp"
-
-namespace ck {
-
-__device__ int32_t llvm_amdgcn_readfirstlane_i32(int32_t i) __asm("llvm.amdgcn.readfirstlane");
-
-} // namespace ck
-#endif
diff --git a/include/ck/utility/amd_wave_read_first_lane.hpp b/include/ck/utility/amd_wave_read_first_lane.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..741b2975af6c5bf99346b1460018eac6fa33b21b
--- /dev/null
+++ b/include/ck/utility/amd_wave_read_first_lane.hpp
@@ -0,0 +1,138 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/utility/functional2.hpp"
+#include "ck/utility/math.hpp"
+
+#include <array>
+#include <cstddef>
+#include <cstdint>
+#include <type_traits>
+
+namespace ck {
+namespace detail {
+
+template <unsigned SizeInBytes>
+struct get_carrier;
+
+template <>
+struct get_carrier<1>
+{
+    using type = uint8_t;
+};
+
+template <>
+struct get_carrier<2>
+{
+    using type = uint16_t;
+};
+
+template <>
+struct get_carrier<3>
+{
+    using type = class carrier
+    {
+        using value_type = uint32_t;
+
+        std::array<std::byte, 3> bytes;
+        static_assert(sizeof(bytes) <= sizeof(value_type));
+
+        // replacement of host std::copy_n()
+        template <typename InputIterator, typename Size, typename OutputIterator>
+        __device__ static OutputIterator copy_n(InputIterator from, Size size, OutputIterator to)
+        {
+            if(0 < size)
+            {
+                *to = *from;
+                ++to;
+                for(Size count = 1; count < size; ++count)
+                {
+                    *to = *++from;
+                    ++to;
+                }
+            }
+
+            return to;
+        }
+
+        // method to trigger template substitution failure
+        __device__ carrier(const carrier& other) noexcept
+        {
+            copy_n(other.bytes.begin(), bytes.size(), bytes.begin());
+        }
+
+        public:
+        __device__ carrier& operator=(value_type value) noexcept
+        {
+            copy_n(reinterpret_cast<const std::byte*>(&value), bytes.size(), bytes.begin());
+
+            return *this;
+        }
+
+        __device__ operator value_type() const noexcept
+        {
+            std::byte result[sizeof(value_type)];
+
+            copy_n(bytes.begin(), bytes.size(), result);
+
+            return *reinterpret_cast<const value_type*>(result);
+        }
+    };
+};
+static_assert(sizeof(get_carrier<3>::type) == 3);
+
+template <>
+struct get_carrier<4>
+{
+    using type = uint32_t;
+};
+
+template <unsigned SizeInBytes>
+using get_carrier_t = typename get_carrier<SizeInBytes>::type;
+
+} // namespace detail
+
+__device__ inline int32_t amd_wave_read_first_lane(int32_t value)
+{
+    return __builtin_amdgcn_readfirstlane(value);
+}
+
+template <
+    typename Object,
+    typename = std::enable_if_t<std::is_class_v<Object> && std::is_trivially_copyable_v<Object>>>
+__device__ auto amd_wave_read_first_lane(const Object& obj)
+{
+    using Size                = unsigned;
+    constexpr Size SgprSize   = 4;
+    constexpr Size ObjectSize = sizeof(Object);
+
+    auto* const from_obj = reinterpret_cast<const std::byte*>(&obj);
+    alignas(Object) std::byte to_obj[ObjectSize];
+
+    constexpr Size RemainedSize             = ObjectSize % SgprSize;
+    constexpr Size CompleteSgprCopyBoundary = ObjectSize - RemainedSize;
+    for(Size offset = 0; offset < CompleteSgprCopyBoundary; offset += SgprSize)
+    {
+        using Sgpr = detail::get_carrier_t<SgprSize>;
+
+        *reinterpret_cast<Sgpr*>(to_obj + offset) =
+            amd_wave_read_first_lane(*reinterpret_cast<const Sgpr*>(from_obj + offset));
+    }
+
+    if constexpr(0 < RemainedSize)
+    {
+        using Carrier = detail::get_carrier_t<RemainedSize>;
+
+        *reinterpret_cast<Carrier*>(to_obj + CompleteSgprCopyBoundary) = amd_wave_read_first_lane(
+            *reinterpret_cast<const Carrier*>(from_obj + CompleteSgprCopyBoundary));
+    }
+
+    /// NOTE: Implicitly start object lifetime. It's better to use std::start_lifetime_at() in this
+    /// scenario
+    return *reinterpret_cast<Object*>(to_obj);
+}
+
+} // namespace ck
diff --git a/include/ck/utility/amd_wmma.hpp b/include/ck/utility/amd_wmma.hpp
index 752876a76920cb401b9334b3fac627e0c0895510..dd7f0b770a11470037b1d33164d830cb9819e8d1 100644
--- a/include/ck/utility/amd_wmma.hpp
+++ b/include/ck/utility/amd_wmma.hpp
@@ -1,14 +1,16 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_AMD_WMMA_HPP
 #define CK_AMD_WMMA_HPP
 
+#include "ck/utility/amd_inline_asm.hpp"
 #include "data_type.hpp"
 // TODO: Add arch limitation
 namespace ck {
 
-// wave32 only
+/********************************WAVE32 MODE***********************************************/
+
 // src: fp16, dst: fp32
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_wmma_f32_16x16x16_f16_w32;
@@ -19,8 +21,18 @@ struct intrin_wmma_f32_16x16x16_f16_w32<16, 16>
     template <class FloatC>
     __device__ static void Run(const half16_t& reg_a, const half16_t& reg_b, FloatC& reg_c)
     {
+        // * Inline assembly need to elimate the duplicated data load, compiler won't help you
+        // delete them.
+        // amd_assembly_wmma_f32_16x16x16_f16_w32(
+        //     reg_a, reg_b, reg_c.template AsType<float8_t>()(Number<0>{}));
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
         reg_c.template AsType<float8_t>()(Number<0>{}) = __builtin_amdgcn_wmma_f32_16x16x16_f16_w32(
             reg_a, reg_b, reg_c.template AsType<float8_t>()[Number<0>{}]);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
     }
 };
 
@@ -34,9 +46,15 @@ struct intrin_wmma_f32_16x16x16_bf16_w32<16, 16>
     template <class FloatC>
     __device__ static void Run(const bhalf16_t& reg_a, const bhalf16_t& reg_b, FloatC& reg_c)
     {
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
         reg_c.template AsType<float8_t>()(Number<0>{}) =
             __builtin_amdgcn_wmma_f32_16x16x16_bf16_w32(
                 reg_a, reg_b, reg_c.template AsType<float8_t>()[Number<0>{}]);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
     }
 };
 
@@ -53,8 +71,14 @@ struct intrin_wmma_f16_16x16x16_f16_w32<16, 16, Opsel>
         // opsel usage
         // false: D0.[0:15] = result
         // true : D0.[16:31]= result
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
         reg_c.template AsType<half16_t>()(Number<0>{}) = __builtin_amdgcn_wmma_f16_16x16x16_f16_w32(
             reg_a, reg_b, reg_c.template AsType<half16_t>()[Number<0>{}], Opsel);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
     }
 };
 
@@ -71,9 +95,15 @@ struct intrin_wmma_bf16_16x16x16_bf16_w32<16, 16, Opsel>
         // opsel usage
         // false: D0.[0:15] = result
         // true : D0.[16:31]= result
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
         reg_c.template AsType<bhalf16_t>()(Number<0>{}) =
             __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w32(
                 reg_a, reg_b, reg_c.template AsType<bhalf16_t>()[Number<0>{}], Opsel);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
     }
 };
 
@@ -87,6 +117,7 @@ struct intrin_wmma_i32_16x16x16_iu8_w32<16, 16, neg_a, neg_b, clamp>
     template <class FloatC>
     __device__ static void Run(const int8x16_t& reg_a, const int8x16_t& reg_b, FloatC& reg_c)
     {
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
         reg_c.template AsType<int32x8_t>()(Number<0>{}) =
             __builtin_amdgcn_wmma_i32_16x16x16_iu8_w32(
                 neg_a,
@@ -95,6 +126,131 @@ struct intrin_wmma_i32_16x16x16_iu8_w32<16, 16, neg_a, neg_b, clamp>
                 bit_cast<int32x4_t>(reg_b),
                 reg_c.template AsType<int32x8_t>()[Number<0>{}],
                 clamp);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
+    }
+};
+
+/********************************WAVE64 MODE***********************************************/
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_wmma_f32_16x16x16_f16_w64;
+
+template <>
+struct intrin_wmma_f32_16x16x16_f16_w64<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const half16_t& reg_a, const half16_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
+        reg_c.template AsType<float4_t>()(Number<0>{}) = __builtin_amdgcn_wmma_f32_16x16x16_f16_w64(
+            reg_a, reg_b, reg_c.template AsType<float4_t>()[Number<0>{}]);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
+    }
+};
+
+// src: bf16, dst: fp32
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_wmma_f32_16x16x16_bf16_w64;
+
+template <>
+struct intrin_wmma_f32_16x16x16_bf16_w64<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const bhalf16_t& reg_a, const bhalf16_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
+        reg_c.template AsType<float4_t>()(Number<0>{}) =
+            __builtin_amdgcn_wmma_f32_16x16x16_bf16_w64(
+                reg_a, reg_b, reg_c.template AsType<float4_t>()[Number<0>{}]);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
+    }
+};
+
+// src: fp16, dst: fp16
+template <index_t MPerWave, index_t NPerWave, index_t Opsel>
+struct intrin_wmma_f16_16x16x16_f16_w64;
+
+template <index_t Opsel>
+struct intrin_wmma_f16_16x16x16_f16_w64<16, 16, Opsel>
+{
+    template <class FloatC>
+    __device__ static void Run(const half16_t& reg_a, const half16_t& reg_b, FloatC& reg_c)
+    {
+        // opsel usage
+        // false: D0.[0:15] = result
+        // true : D0.[16:31]= result
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
+        reg_c.template AsType<half8_t>()(Number<0>{}) = __builtin_amdgcn_wmma_f16_16x16x16_f16_w64(
+            reg_a, reg_b, reg_c.template AsType<half8_t>()[Number<0>{}], Opsel);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
+    }
+};
+
+// src: bf16, dst: bf16
+template <index_t MPerWave, index_t NPerWave, index_t Opsel>
+struct intrin_wmma_bf16_16x16x16_bf16_w64;
+
+template <index_t Opsel>
+struct intrin_wmma_bf16_16x16x16_bf16_w64<16, 16, Opsel>
+{
+    template <class FloatC>
+    __device__ static void Run(const bhalf16_t& reg_a, const bhalf16_t& reg_b, FloatC& reg_c)
+    {
+        // opsel usage
+        // false: D0.[0:15] = result
+        // true : D0.[16:31]= result
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
+        reg_c.template AsType<bhalf8_t>()(Number<0>{}) =
+            __builtin_amdgcn_wmma_bf16_16x16x16_bf16_w64(
+                reg_a, reg_b, reg_c.template AsType<bhalf8_t>()[Number<0>{}], Opsel);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
+    }
+};
+
+// src: iu8, dst: i32
+template <index_t MPerWave, index_t NPerWave, bool neg_a, bool neg_b, bool clamp>
+struct intrin_wmma_i32_16x16x16_iu8_w64;
+
+template <bool neg_a, bool neg_b, bool clamp>
+struct intrin_wmma_i32_16x16x16_iu8_w64<16, 16, neg_a, neg_b, clamp>
+{
+    template <class FloatC>
+    __device__ static void Run(const int8x16_t& reg_a, const int8x16_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
+        reg_c.template AsType<int32x4_t>()(Number<0>{}) =
+            __builtin_amdgcn_wmma_i32_16x16x16_iu8_w64(
+                neg_a,
+                bit_cast<int32x4_t>(reg_a),
+                neg_b,
+                bit_cast<int32x4_t>(reg_b),
+                reg_c.template AsType<int32x4_t>()[Number<0>{}],
+                clamp);
+#else
+        ignore = reg_a;
+        ignore = reg_b;
+        ignore = reg_c;
+#endif
     }
 };
 
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index b4be0cbee70cbe0e6f6827b9c1f74fb9125e2a39..ea7755036fc64b0af6aa340e062e1277c485e785 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_AMD_XDLOPS_HPP
 #define CK_AMD_XDLOPS_HPP
@@ -297,6 +297,44 @@ struct intrin_mfma_i32_16x16x16i8<16, 16>
     }
 };
 
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_i32_32x32x16i8;
+
+template <>
+struct intrin_mfma_i32_32x32x16i8<32, 32>
+{
+    template <class FloatC>
+    __device__ static void Run(const int8x8_t& reg_a, const int8x8_t& reg_b, FloatC& reg_c)
+    {
+        reg_c.template AsType<int32x16_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_i32_32x32x16_i8(bit_cast<int64_t>(reg_a),
+                                                  bit_cast<int64_t>(reg_b),
+                                                  reg_c.template AsType<int32x16_t>()[Number<0>{}],
+                                                  0,
+                                                  0,
+                                                  0);
+    }
+};
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_i32_16x16x32i8;
+
+template <>
+struct intrin_mfma_i32_16x16x32i8<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const int8x8_t& reg_a, const int8x8_t& reg_b, FloatC& reg_c)
+    {
+        reg_c.template AsType<int32x4_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_i32_16x16x32i8(bit_cast<int64_t>(reg_a),
+                                                 bit_cast<int64_t>(reg_b),
+                                                 reg_c.template AsType<int32x4_t>()[Number<0>{}],
+                                                 0,
+                                                 0,
+                                                 0);
+    }
+};
+
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f64_16x16x4f64;
 
@@ -306,7 +344,7 @@ struct intrin_mfma_f64_16x16x4f64<16, 16>
     template <class FloatC>
     __device__ static void Run(const double& reg_a, const double& reg_b, FloatC& reg_c)
     {
-#ifdef __gfx90a__
+#if defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
         reg_c.template AsType<double4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f64_16x16x4f64(
             reg_a, reg_b, reg_c.template AsType<double4_t>()[Number<0>{}], 0, 0, 0);
 #else
@@ -316,5 +354,68 @@ struct intrin_mfma_f64_16x16x4f64<16, 16>
 #endif
     }
 };
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_32x32x16f8f8;
+
+template <>
+struct intrin_mfma_f32_32x32x16f8f8<32, 32>
+{
+    template <class FloatC>
+    __device__ static void Run(const f8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float16_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_f32_32x32x16_fp8_fp8(
+                bit_cast<long>(reg_a),
+                bit_cast<long>(reg_b),
+                reg_c.template AsType<float16_t>()[Number<0>{}],
+                0,
+                0,
+                0);
+#else
+        vector_type<f8_t, 8> reg_a_v(reg_a);
+        vector_type<f8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<f8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<f8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_32x32x2f32<32, 32>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_16x16x32f8f8;
+
+template <>
+struct intrin_mfma_f32_16x16x32f8f8<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const f8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f32_16x16x32_fp8_fp8(
+            bit_cast<long>(reg_a),
+            bit_cast<long>(reg_b),
+            reg_c.template AsType<float4_t>()[Number<0>{}],
+            0,
+            0,
+            0);
+#else
+        vector_type<f8_t, 8> reg_a_v(reg_a);
+        vector_type<f8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<f8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<f8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_16x16x4f32<16, 16>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
 } // namespace ck
 #endif
diff --git a/include/ck/utility/array.hpp b/include/ck/utility/array.hpp
index 370a457fe9d97a621e6f91e78f14733398ebf756..f63ce5e5a07a796888cb60ae8da0c855df75e7ff 100644
--- a/include/ck/utility/array.hpp
+++ b/include/ck/utility/array.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_ARRAY_HPP
 #define CK_ARRAY_HPP
diff --git a/include/ck/utility/array_multi_index.hpp b/include/ck/utility/array_multi_index.hpp
index 9b8d5b95e9f6241b574104313b5b8fa2984ab18d..c0c1ea65fc77dd625aaaf6456a6628e09df27b51 100644
--- a/include/ck/utility/array_multi_index.hpp
+++ b/include/ck/utility/array_multi_index.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_ARRAY_MULTI_INDEX_HPP
 #define CK_ARRAY_MULTI_INDEX_HPP
diff --git a/include/ck/utility/c_style_pointer_cast.hpp b/include/ck/utility/c_style_pointer_cast.hpp
index 6e8b0081587afcb2db159a05ecd5fb940def68ff..610e393a77216500448c3682b5db9b9860a077da 100644
--- a/include/ck/utility/c_style_pointer_cast.hpp
+++ b/include/ck/utility/c_style_pointer_cast.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_C_STYLE_POINTER_CAST_HPP
 #define CK_C_STYLE_POINTER_CAST_HPP
diff --git a/include/ck/utility/common_header.hpp b/include/ck/utility/common_header.hpp
index 1378bbe448e2e1862fed04596fba7f4b106d03b7..f95660a8a47d2c93a65a497a22a74c4c0b6eaea4 100644
--- a/include/ck/utility/common_header.hpp
+++ b/include/ck/utility/common_header.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -24,6 +24,7 @@
 #include "ck/utility/tuple.hpp"
 #include "ck/utility/tuple_helper.hpp"
 #include "ck/utility/type.hpp"
+#include "ck/utility/type_convert.hpp"
 #include "ck/utility/magic_division.hpp"
 #include "ck/utility/c_style_pointer_cast.hpp"
 #include "ck/utility/is_known_at_compile_time.hpp"
@@ -33,6 +34,7 @@
 #include "ck/utility/debug.hpp"
 
 #include "ck/utility/amd_buffer_addressing.hpp"
+#include "ck/utility/amd_wave_read_first_lane.hpp"
 #include "ck/utility/generic_memory_space_atomic.hpp"
 #include "ck/utility/get_id.hpp"
 #include "ck/utility/thread_group.hpp"
diff --git a/include/ck/utility/container_element_picker.hpp b/include/ck/utility/container_element_picker.hpp
index abc5185e04a5079edb95fe15bfcd6788256d384a..838147e420cc7c95b63273b32ae3b2be2429c017 100644
--- a/include/ck/utility/container_element_picker.hpp
+++ b/include/ck/utility/container_element_picker.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_CONTAINER_ELEMENT_PICKER_HPP
 #define CK_CONTAINER_ELEMENT_PICKER_HPP
diff --git a/include/ck/utility/container_helper.hpp b/include/ck/utility/container_helper.hpp
index c8b02bc5acaf8542bf028fff39c42061db3b3296..9c7b954565d386a8fdecd21052b102e750ab7102 100644
--- a/include/ck/utility/container_helper.hpp
+++ b/include/ck/utility/container_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_CONTAINER_HELPER_HPP
 #define CK_CONTAINER_HELPER_HPP
diff --git a/include/ck/utility/data_type.hpp b/include/ck/utility/data_type.hpp
index c1306c40e95def0c453f81f9a6245fbbfcb49891..a899873847c6aeaf57527e45f1e0975b559a7da8 100644
--- a/include/ck/utility/data_type.hpp
+++ b/include/ck/utility/data_type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -12,6 +12,7 @@ using half_t  = _Float16;
 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 using int4_t = _BitInt(4);
 #endif
+using f8_t = uint8_t;
 
 // vector_type
 template <typename T, index_t N>
@@ -142,6 +143,13 @@ struct scalar_type<int4_t>
 };
 #endif
 
+template <>
+struct scalar_type<f8_t>
+{
+    using type                           = f8_t;
+    static constexpr index_t vector_size = 1;
+};
+
 //
 template <typename T>
 struct vector_type<T, 1>
@@ -898,6 +906,8 @@ struct vector_type<T, 256>
     }
 };
 
+using int64_t = long;
+
 // fp64
 using double2_t = typename vector_type<double, 2>::type;
 using double4_t = typename vector_type<double, 4>::type;
@@ -942,107 +952,13 @@ using int8x16_t = typename vector_type<int8_t, 16>::type;
 using int8x32_t = typename vector_type<int8_t, 32>::type;
 using int8x64_t = typename vector_type<int8_t, 64>::type;
 
-// Convert X to Y
-template <typename Y, typename X>
-__host__ __device__ constexpr Y type_convert(X x)
-{
-    static_assert(!std::is_reference_v<Y> && !std::is_reference_v<X>);
-
-    return static_cast<Y>(x);
-}
-
-// convert bfp16 to fp32
-template <>
-inline __host__ __device__ constexpr float type_convert<float, bhalf_t>(bhalf_t x)
-{
-    union
-    {
-        uint32_t int32;
-        float fp32;
-    } u = {uint32_t(x) << 16};
-
-    return u.fp32;
-}
-
-// convert fp32 to bfp16
-template <>
-inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, float>(float x)
-{
-    union
-    {
-        float fp32;
-        uint32_t int32;
-    } u = {x};
-
-    // When the exponent bits are not all 1s, then the value is zero, normal,
-    // or subnormal. We round the bfloat16 mantissa up by adding 0x7FFF, plus
-    // 1 if the least significant bit of the bfloat16 mantissa is 1 (odd).
-    // This causes the bfloat16's mantissa to be incremented by 1 if the 16
-    // least significant bits of the float mantissa are greater than 0x8000,
-    // or if they are equal to 0x8000 and the least significant bit of the
-    // bfloat16 mantissa is 1 (odd). This causes it to be rounded to even when
-    // the lower 16 bits are exactly 0x8000. If the bfloat16 mantissa already
-    // has the value 0x7f, then incrementing it causes it to become 0x00 and
-    // the exponent is incremented by one, which is the next higher FP value
-    // to the unrounded bfloat16 value. When the bfloat16 value is subnormal
-    // with an exponent of 0x00 and a mantissa of 0x7f, it may be rounded up
-    // to a normal value with an exponent of 0x01 and a mantissa of 0x00.
-    // When the bfloat16 value has an exponent of 0xFE and a mantissa of 0x7F,
-    // incrementing it causes it to become an exponent of 0xFF and a mantissa
-    // of 0x00, which is Inf, the next higher value to the unrounded value.
-    bool flag0 = ~u.int32 & 0x7f800000;
-
-    // When all of the exponent bits are 1, the value is Inf or NaN.
-    // Inf is indicated by a zero mantissa. NaN is indicated by any nonzero
-    // mantissa bit. Quiet NaN is indicated by the most significant mantissa
-    // bit being 1. Signaling NaN is indicated by the most significant
-    // mantissa bit being 0 but some other bit(s) being 1. If any of the
-    // lower 16 bits of the mantissa are 1, we set the least significant bit
-    // of the bfloat16 mantissa, in order to preserve signaling NaN in case
-    // the bfloat16's mantissa bits are all 0.
-    bool flag1 = !flag0 && (u.int32 & 0xffff);
-
-    u.int32 += flag0 ? 0x7fff + ((u.int32 >> 16) & 1) : 0; // Round to nearest, round to even
-    u.int32 |= flag1 ? 0x10000 : 0x0;                      // Preserve signaling NaN
-
-    return uint16_t(u.int32 >> 16);
-}
-
-// convert fp16 to bf16
-template <>
-inline __host__ __device__ bhalf_t type_convert<bhalf_t, half_t>(half_t x)
-{
-    union
-    {
-        float fp32;
-        uint32_t int32;
-    } u = {static_cast<float>(x)};
-
-    return uint16_t(u.int32 >> 16);
-}
-
-template <>
-inline __host__ __device__ bhalf2_t type_convert<bhalf2_t, half2_t>(half2_t x)
-{
-    float y0{0}, y1{0};
-    bhalf2_t y{0};
-    asm volatile("\n \
-            v_cvt_f32_f16 %0, %1 \n \
-            "
-                 : "=v"(y0)
-                 : "v"(x));
-    asm volatile("\n \
-            v_cvt_f32_f16 %0, %1 src0_sel:WORD_1\n \
-            "
-                 : "=v"(y1)
-                 : "v"(x));
-    asm volatile("\n \
-            v_pack_b32_f16 %0, %1, %2 op_sel:[1, 1] \n \
-            "
-                 : "=v"(y)
-                 : "v"(y0), "v"(y1));
-    return y;
-}
+// f8
+using f8x2_t  = typename vector_type<f8_t, 2>::type;
+using f8x4_t  = typename vector_type<f8_t, 4>::type;
+using f8x8_t  = typename vector_type<f8_t, 8>::type;
+using f8x16_t = typename vector_type<f8_t, 16>::type;
+using f8x32_t = typename vector_type<f8_t, 32>::type;
+using f8x64_t = typename vector_type<f8_t, 64>::type;
 
 template <typename T>
 struct NumericLimits
@@ -1090,6 +1006,23 @@ struct NumericLimits<int4_t>
 };
 #endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 
+template <>
+struct NumericLimits<f8_t>
+{
+    static constexpr uint8_t binary_min    = 0x08; // 0b00001000
+    static constexpr uint8_t binary_max    = 0x77; // 0b01110111
+    static constexpr uint8_t binary_lowest = 0xF7; // 0b11110111
+    static constexpr uint8_t binary_qnan   = 0x80; // 0b10000000
+
+    __host__ __device__ static constexpr f8_t Min() { return bit_cast<f8_t>(binary_min); }
+
+    __host__ __device__ static constexpr f8_t Max() { return bit_cast<f8_t>(binary_max); }
+
+    __host__ __device__ static constexpr f8_t Lowest() { return bit_cast<f8_t>(binary_lowest); }
+
+    __host__ __device__ static constexpr f8_t QuietNaN() { return bit_cast<f8_t>(binary_qnan); }
+};
+
 } // namespace ck
 
 namespace std {
diff --git a/include/ck/utility/debug.hpp b/include/ck/utility/debug.hpp
index 593bbb711672f7dc6e3db22d877bf249b5dff56d..80346f0d9f6f9e5a6a28dcab4e8a7666639394ca 100644
--- a/include/ck/utility/debug.hpp
+++ b/include/ck/utility/debug.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef UTILITY_DEBUG_HPP
 #define UTILITY_DEBUG_HPP
diff --git a/include/ck/utility/dynamic_buffer.hpp b/include/ck/utility/dynamic_buffer.hpp
index 114444d0714c68d45f5d2735aefaa826c7b783da..9009694f9774f115b687d10ba9cf60db208c7282 100644
--- a/include/ck/utility/dynamic_buffer.hpp
+++ b/include/ck/utility/dynamic_buffer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -19,7 +19,8 @@ namespace ck {
 template <AddressSpaceEnum BufferAddressSpace,
           typename T,
           typename ElementSpaceSize,
-          bool InvalidElementUseNumericalZeroValue>
+          bool InvalidElementUseNumericalZeroValue,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 struct DynamicBuffer
 {
     using type = T;
@@ -77,13 +78,16 @@ struct DynamicBuffer
 
             if constexpr(InvalidElementUseNumericalZeroValue)
             {
-                return amd_buffer_load_invalid_element_return_zero<remove_cvref_t<T>, t_per_x>(
+                return amd_buffer_load_invalid_element_return_zero<remove_cvref_t<T>,
+                                                                   t_per_x,
+                                                                   coherence>(
                     p_data_, i, is_valid_element, element_space_size_);
             }
             else
             {
                 return amd_buffer_load_invalid_element_return_customized_value<remove_cvref_t<T>,
-                                                                               t_per_x>(
+                                                                               t_per_x,
+                                                                               coherence>(
                     p_data_, i, is_valid_element, element_space_size_, invalid_element_value_);
             }
         }
@@ -183,7 +187,7 @@ struct DynamicBuffer
         {
             constexpr index_t t_per_x = scalar_per_x_vector / scalar_per_t_vector;
 
-            amd_buffer_store<remove_cvref_t<T>, t_per_x>(
+            amd_buffer_store<remove_cvref_t<T>, t_per_x, coherence>(
                 x, p_data_, i, is_valid_element, element_space_size_);
         }
         else if constexpr(GetAddressSpace() == AddressSpaceEnum::Lds &&
@@ -388,14 +392,19 @@ struct DynamicBuffer
     __host__ __device__ static constexpr bool IsDynamicBuffer() { return true; }
 };
 
-template <AddressSpaceEnum BufferAddressSpace, typename T, typename ElementSpaceSize>
+template <AddressSpaceEnum BufferAddressSpace,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence,
+          typename T,
+          typename ElementSpaceSize>
 __host__ __device__ constexpr auto make_dynamic_buffer(T* p, ElementSpaceSize element_space_size)
 {
-    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, true>{p, element_space_size};
+    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, true, coherence>{
+        p, element_space_size};
 }
 
 template <
     AddressSpaceEnum BufferAddressSpace,
+    AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence,
     typename T,
     typename ElementSpaceSize,
     typename X,
@@ -403,7 +412,7 @@ template <
 __host__ __device__ constexpr auto
 make_dynamic_buffer(T* p, ElementSpaceSize element_space_size, X invalid_element_value)
 {
-    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, false>{
+    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, false, coherence>{
         p, element_space_size, invalid_element_value};
 }
 
diff --git a/include/ck/utility/enable_if.hpp b/include/ck/utility/enable_if.hpp
index 297434b0dddd5f1a680176c3bd099bedfda8dff4..c0a3c99f1fdafea9f151fe9fc319c2f7aaa0ffda 100644
--- a/include/ck/utility/enable_if.hpp
+++ b/include/ck/utility/enable_if.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/f8_utils.hpp b/include/ck/utility/f8_utils.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..bb13f98154ef4ec072529c26ea993f8e1a444abf
--- /dev/null
+++ b/include/ck/utility/f8_utils.hpp
@@ -0,0 +1,250 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+
+namespace ck {
+
+// fp8 rounding modes
+// use standard for rounding to nearest, the faster one
+// use stochastic for stochastic rounding, helps to avoid error accumulation
+enum class f8_rounding_mode
+{
+    standard,
+    stochastic
+};
+
+} // namespace ck
+
+namespace ck::utils {
+
+namespace {
+
+template <typename T, bool negative_zero_nan, bool clip, bool stoch>
+__host__ __device__ f8_t run_cast_to_f8(T x, uint32_t rng)
+{
+    // check data type
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+
+    // fp8 exponent/mantissa layout
+    constexpr int f8_exp  = 4;
+    constexpr int f8_mant = 3;
+
+    // resulting type exponent/mantissa layout
+    constexpr int type_exp  = is_half ? 5 : 8;
+    constexpr int type_mant = is_half ? 10 : 23;
+
+    int exponent;
+    uint32_t head, mantissa, sign;
+    // nan code is same for float and half
+    constexpr uint8_t nan_code  = 0x80;
+    constexpr uint32_t nan_mask = is_half ? 0x7C00 : 0x7F800000;
+
+    // convert to bitwise
+    typedef typename std::conditional<std::is_same<T, half_t>::value, uint16_t, uint32_t>::type
+        T_bitwise;
+    T_bitwise x_bitwise = *(reinterpret_cast<T_bitwise*>(&x));
+
+    // unpack the input, depends on datatype
+    if constexpr(is_float)
+    {
+        head     = x_bitwise & 0xFF800000;
+        mantissa = x_bitwise & 0x7FFFFF;
+        exponent = (head >> type_mant) & 0xFF;
+        sign     = head >> (type_exp + type_mant);
+    }
+    else if constexpr(is_half)
+    {
+        head     = x_bitwise & 0xFC00;
+        mantissa = x_bitwise & 0x3FF;
+        exponent = (head >> type_mant) & 0x1F;
+        sign     = head >> (type_exp + type_mant);
+    }
+
+    uint32_t signed_inf   = (sign << (type_exp + type_mant)) + (((1 << type_exp) - 1) << type_mant);
+    uint32_t drop_mask    = (1 << (type_mant - f8_mant)) - 1;
+    constexpr int max_exp = (1 << f8_exp) - (negative_zero_nan ? 1 : 2);
+    constexpr int exp_low_cutoff =
+        (1 << (type_exp - 1)) - (1 << (f8_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
+
+    if constexpr(negative_zero_nan)
+    {
+        if((x_bitwise & nan_mask) == nan_mask)
+            return nan_code;
+    }
+    else
+    {
+        if((x_bitwise & nan_mask) == nan_mask)
+            return signed_inf + (mantissa != 0 ? 1 : 0);
+    }
+
+    // check if x is 0.0
+    if(x_bitwise == 0)
+        return 0;
+
+    exponent -= exp_low_cutoff - 1;
+    if(exponent <= 0)
+        drop_mask = (1 << (type_mant - f8_mant + 1 - exponent)) - 1;
+    mantissa += 1 << type_mant;
+    // apply random number if needed
+    mantissa += (stoch ? rng : mantissa) & drop_mask;
+    if(mantissa >= (2 << type_mant))
+    {
+        mantissa >>= 1;
+        exponent++;
+    }
+    mantissa >>= (type_mant - f8_mant);
+
+    // check negative exponent
+    if(exponent <= 0)
+    {
+        if(x_bitwise == 0)
+            return 0;
+        else
+        {
+            // subnormal range; represented by a subnormal float8 (exponent 0)
+            // and involves loss of accuracy
+            mantissa >>= 1 - exponent;
+            exponent = 0;
+        }
+    }
+    // above range: quantize to maximum possible float of the same sign
+    else if(exponent > max_exp)
+    {
+        if(clip)
+        {
+            mantissa = (1 << f8_mant) - 1;
+            exponent = max_exp;
+        }
+        else
+        {
+            return signed_inf;
+        }
+    }
+
+    // check if x is 0.0 or -0.0
+    if(exponent == 0 && mantissa == 0)
+        return negative_zero_nan ? 0 : (sign << (f8_exp + f8_mant));
+    mantissa &= (1 << f8_mant) - 1;
+    return (sign << (f8_exp + f8_mant)) | (exponent << f8_mant) | mantissa;
+}
+
+template <typename T, bool negative_zero_nan>
+__host__ __device__ T run_cast_from_f8(f8_t x)
+{
+    // check data type
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+
+    // fp8 exponent/mantissa layout
+    constexpr int f8_exp  = 4;
+    constexpr int f8_mant = 3;
+
+    // resulting type exponent/mantissa layout
+    constexpr int type_exp  = is_half ? 5 : 8;
+    constexpr int type_mant = is_half ? 10 : 23;
+
+    // prepare the codes
+    constexpr uint8_t nan_code = 0x80;
+    T fInf, fNegInf, fNaN, fNeg0;
+    if constexpr(is_half)
+    {
+        constexpr uint16_t ihInf    = 0x7C00;
+        constexpr uint16_t ihNegInf = 0xFC00;
+        constexpr uint16_t ihNaN    = 0x7C01;
+        constexpr uint16_t ihNeg0   = 0x8000;
+        fInf                        = *(reinterpret_cast<const half_t*>(&ihInf));
+        fNegInf                     = *(reinterpret_cast<const half_t*>(&ihNegInf));
+        fNaN                        = *(reinterpret_cast<const half_t*>(&ihNaN));
+        fNeg0                       = *(reinterpret_cast<const half_t*>(&ihNeg0));
+    }
+    else if constexpr(is_float)
+    {
+        constexpr uint32_t ifInf    = 0x7F800000;
+        constexpr uint32_t ifNegInf = 0xFF800000;
+        constexpr uint32_t ifNaN    = 0x7F800001;
+        constexpr uint32_t ifNeg0   = 0x80000000;
+        fInf                        = *(reinterpret_cast<const float*>(&ifInf));
+        fNegInf                     = *(reinterpret_cast<const float*>(&ifNegInf));
+        fNaN                        = *(reinterpret_cast<const float*>(&ifNaN));
+        fNeg0                       = *(reinterpret_cast<const float*>(&ifNeg0));
+    }
+
+    // unpack the input
+    uint32_t sign     = x >> (f8_exp + f8_mant);
+    uint32_t mantissa = x & ((1 << f8_mant) - 1);
+    int exponent      = (x & 0x7F) >> f8_mant;
+
+    constexpr int exp_low_cutoff =
+        (1 << (type_exp - 1)) - (1 << (f8_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
+    typename std::conditional<std::is_same<T, half_t>::value, uint16_t, uint32_t>::type retval;
+
+    if constexpr(negative_zero_nan)
+    {
+        if(x == nan_code)
+            return fNaN;
+    }
+    else
+    {
+        if(x == nan_code)
+            return fNeg0;
+        if(exponent == ((1 << f8_exp) - 1))
+            return (mantissa == 0) ? (sign ? fNegInf : fInf) : fNaN;
+    }
+
+    // subnormal input
+    if(exponent == 0)
+    {
+        // guaranteed mantissa!=0 since cases 0x0 and 0x80 are handled above
+        int sh = 1 + __builtin_clz(mantissa) - ((1 + type_exp + type_mant) - f8_mant);
+        mantissa <<= sh;
+        mantissa &= ((1 << f8_mant) - 1);
+        exponent += 1 - sh;
+    }
+    exponent += exp_low_cutoff - 1;
+    mantissa <<= type_mant - f8_mant;
+
+    // subnormal output (occurs when T=half, we=5, negative_zero_nan=true)
+    if(exponent <= 0)
+    {
+        mantissa |= 1 << type_mant;
+        mantissa >>= 1 - exponent;
+        exponent = 0;
+    }
+
+    retval = (sign << (type_exp + type_mant)) | (exponent << type_mant) | mantissa;
+    return *(reinterpret_cast<const T*>(&retval));
+}
+
+} // namespace
+
+template <typename T, bool negative_zero_nan, bool clip, bool stoch>
+__host__ __device__ f8_t cast_to_f8(T x, uint32_t rng)
+{
+    // check datatype
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+    static_assert(is_half || is_float, "Only half and float can be casted to f8.");
+
+    return run_cast_to_f8<T, negative_zero_nan, clip, stoch>(x, rng);
+}
+
+template <typename T, bool negative_zero_nan>
+__host__ __device__ T cast_from_f8(f8_t x)
+{
+    // check datatype
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+    static_assert(is_half || is_float, "only half and float are supported.");
+
+    // check if x is 0.0
+    if(x == 0)
+        return static_cast<T>(0);
+
+    return run_cast_from_f8<T, negative_zero_nan>(x);
+}
+
+} // namespace ck::utils
diff --git a/include/ck/utility/functional.hpp b/include/ck/utility/functional.hpp
index 08e730782f386cf5788c64bc04a008dc6cb37b28..91797d24092e3e32ad4a6bd40958952b124d9978 100644
--- a/include/ck/utility/functional.hpp
+++ b/include/ck/utility/functional.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/functional2.hpp b/include/ck/utility/functional2.hpp
index 6f125ca4c944777d02ae1083334bec4602ad68c4..99c65f4eb85b67231557b1916fae10c6568e676d 100644
--- a/include/ck/utility/functional2.hpp
+++ b/include/ck/utility/functional2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/functional3.hpp b/include/ck/utility/functional3.hpp
index 06b67ef7e3fdec25a1ab2e18e23c904342233ebe..97605a7adeb8ae64e4e5a32debe9386295923b6c 100644
--- a/include/ck/utility/functional3.hpp
+++ b/include/ck/utility/functional3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/functional4.hpp b/include/ck/utility/functional4.hpp
index 6eeaf15c9b7ac283f7fcac96c195d28f179b6065..b5f3df8d7c517dfaf01320e41721da174883c2d9 100644
--- a/include/ck/utility/functional4.hpp
+++ b/include/ck/utility/functional4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_FUNCTIONAL4_HPP
 #define CK_FUNCTIONAL4_HPP
diff --git a/include/ck/utility/generic_memory_space_atomic.hpp b/include/ck/utility/generic_memory_space_atomic.hpp
index 1dacf7dda0c29f9906db28bc1c1f5982f7f801f2..e7d1814eb930a45a42b95f40ffac8e688f07cec3 100644
--- a/include/ck/utility/generic_memory_space_atomic.hpp
+++ b/include/ck/utility/generic_memory_space_atomic.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "data_type.hpp"
diff --git a/include/ck/utility/get_id.hpp b/include/ck/utility/get_id.hpp
index 44ff438155d2a9c9ba9b0925d4e9fe95c1fa6bce..77564c6130baf45bfd331e69fc437fb7c7c96d18 100644
--- a/include/ck/utility/get_id.hpp
+++ b/include/ck/utility/get_id.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/get_shift.hpp b/include/ck/utility/get_shift.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0a93081cfd01ce41cf305e4d69e127053de788cc
--- /dev/null
+++ b/include/ck/utility/get_shift.hpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+namespace ck {
+
+template <index_t N>
+static constexpr __device__ index_t get_shift()
+{
+    return (get_shift<N / 2>() + 1);
+};
+
+template <>
+constexpr __device__ index_t get_shift<1>()
+{
+    return (0);
+}
+
+} // namespace ck
diff --git a/include/ck/utility/ignore.hpp b/include/ck/utility/ignore.hpp
index ac33cbf9a508f5e6402d17d6285ae34a8196f6eb..f70a182fd4e5c6cc623acb2c65b5551b2a1acd14 100644
--- a/include/ck/utility/ignore.hpp
+++ b/include/ck/utility/ignore.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/inner_product.hpp b/include/ck/utility/inner_product.hpp
index 0f45ec177acf6d053b50beb928c9569bf67d210c..b13bccb5ab3fd4e1a01470aee97bca78d93e286c 100644
--- a/include/ck/utility/inner_product.hpp
+++ b/include/ck/utility/inner_product.hpp
@@ -1,8 +1,9 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "data_type.hpp"
+#include "type_convert.hpp"
 
 namespace ck {
 
@@ -135,6 +136,28 @@ __device__ void inner_product<half8_t, half8_t, float>(const half8_t& a, const h
                   c);
 }
 
+template <>
+__device__ void inner_product<int8_t, int8_t, int32_t>(const int8_t& a, const int8_t& b, int32_t& c)
+{
+    c += type_convert<int32_t>(a) * type_convert<int32_t>(b);
+}
+
+template <>
+__device__ void
+inner_product<int8x2_t, int8x2_t, int32_t>(const int8x2_t& a, const int8x2_t& b, int32_t& c)
+{
+    constexpr auto I0 = Number<0>{};
+    constexpr auto I1 = Number<1>{};
+
+    inner_product(vector_type<int8_t, 2>{a}.AsType<int8_t>()[I0],
+                  vector_type<int8_t, 2>{b}.AsType<int8_t>()[I0],
+                  c);
+
+    inner_product(vector_type<int8_t, 2>{a}.AsType<int8_t>()[I1],
+                  vector_type<int8_t, 2>{b}.AsType<int8_t>()[I1],
+                  c);
+}
+
 template <>
 __device__ void
 inner_product<int8x4_t, int8x4_t, int32_t>(const int8x4_t& a, const int8x4_t& b, int32_t& c)
diff --git a/include/ck/utility/integral_constant.hpp b/include/ck/utility/integral_constant.hpp
index 9aab4e24214a884bdfb391e0f9040bb3af2630ec..376070eb3d8ac326603b71e52e76949c168f4219 100644
--- a/include/ck/utility/integral_constant.hpp
+++ b/include/ck/utility/integral_constant.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/is_known_at_compile_time.hpp b/include/ck/utility/is_known_at_compile_time.hpp
index 8198154422e5d5228b79c6539bddcd5070d1e25c..2cafc3e6f2fafd247e378446a15a7e16c019c914 100644
--- a/include/ck/utility/is_known_at_compile_time.hpp
+++ b/include/ck/utility/is_known_at_compile_time.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/magic_division.hpp b/include/ck/utility/magic_division.hpp
index a5e8e9216519074d6542d379512c5b8d13ec21c9..f19030d4e9ca38971f3ffbfd5e4ff6b951eb16b3 100644
--- a/include/ck/utility/magic_division.hpp
+++ b/include/ck/utility/magic_division.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/math.hpp b/include/ck/utility/math.hpp
index 397f8783735cfc9b112c11c5af5a0d086f1a9450..804c4254c2d21f0cb6d2501a3f9c476def22025e 100644
--- a/include/ck/utility/math.hpp
+++ b/include/ck/utility/math.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -168,6 +168,10 @@ __device__ double exp<double>(double x)
     return exp(x);
 }
 
+static inline __host__ float exp(float x) { return std::expf(x); }
+
+static inline __host__ double exp(double x) { return std::exp(x); }
+
 // disallow implicit type casting
 template <typename T>
 __device__ T log(T x);
diff --git a/include/ck/utility/math_v2.hpp b/include/ck/utility/math_v2.hpp
index 73347032414d42b3c691d907da4b957a1f72d921..1cac2cc0c7172b4b1915845d6c7f3fa853653683 100644
--- a/include/ck/utility/math_v2.hpp
+++ b/include/ck/utility/math_v2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -83,10 +83,24 @@ static inline __host__ bool isnan(int4_t x)
 };
 #endif
 
+static inline __host__ half_t sqrt(half_t x)
+{
+    return static_cast<half_t>(std::sqrt(static_cast<float>(x)));
+};
+
 static inline __host__ float sqrt(float x) { return std::sqrt(x); };
 
 static inline __host__ double sqrt(double x) { return std::sqrt(x); };
 
+static inline __host__ half_t tanh(half_t x)
+{
+    return static_cast<half_t>(std::tanh(static_cast<float>(x)));
+};
+
+static inline __host__ float tanh(float x) { return std::tanh(x); };
+
+static inline __host__ double tanh(double x) { return std::tanh(x); };
+
 // math functions for the HIP kernel,  some are implemented by calling hip builtin functions
 
 static inline __device__ float abs(float x) { return ::abs(x); };
@@ -158,9 +172,23 @@ static inline __device__ bool isnan(half_t x)
     return (xx & 0x7FFF) > 0x7C00;
 };
 
-static inline __device__ float sqrt(float x) { return ::sqrtf(x); };
+static inline __device__ half_t sqrt(half_t x)
+{
+    return static_cast<half_t>(__builtin_amdgcn_sqrtf(static_cast<float>(x)));
+};
+
+static inline __device__ float sqrt(float x) { return __builtin_amdgcn_sqrtf(x); };
+
+static inline __device__ double sqrt(double x) { return __builtin_amdgcn_sqrt(x); };
+
+static inline __device__ half_t tanh(half_t x)
+{
+    return static_cast<half_t>(::tanhf(static_cast<float>(x)));
+};
+
+static inline __device__ float tanh(float x) { return ::tanhf(x); };
 
-static inline __device__ double sqrt(double x) { return ::sqrt(x); };
+static inline __device__ double tanh(double x) { return ::tanh(x); };
 
 } // namespace math
 } // namespace ck
diff --git a/include/ck/utility/multi_index.hpp b/include/ck/utility/multi_index.hpp
index 1d544c0906cae3c61f4d7ce27e74e7636c3919b5..9f7ba8bff63b3f6e625b98f66accd875ee118594 100644
--- a/include/ck/utility/multi_index.hpp
+++ b/include/ck/utility/multi_index.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/number.hpp b/include/ck/utility/number.hpp
index f3ca6b61dc6ac08330a6cf1633148bb9bad8cd81..d29afd31a7ef03ef3fdbf11f8e9a88132e07fe82 100644
--- a/include/ck/utility/number.hpp
+++ b/include/ck/utility/number.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_NUMBER_HPP
 #define CK_NUMBER_HPP
diff --git a/include/ck/utility/print.hpp b/include/ck/utility/print.hpp
deleted file mode 100644
index eed1ca42c7346e32f4d6924452d568bf431d6c24..0000000000000000000000000000000000000000
--- a/include/ck/utility/print.hpp
+++ /dev/null
@@ -1,25 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_PRINT_HPP
-#define CK_PRINT_HPP
-
-#include "array.hpp"
-#include "statically_indexed_array.hpp"
-#include "container_helper.hpp"
-#include "sequence.hpp"
-
-namespace ck {
-
-template <typename T>
-__host__ __device__ void print_array(const char* s, T a)
-{
-    constexpr index_t nsize = a.Size();
-
-    printf("%s size %d, {", s, nsize);
-    static_for<0, nsize, 1>{}([&a](auto i) constexpr { printf("%d, ", int32_t{a[i]}); });
-    printf("}\n");
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/utility/random_gen.hpp b/include/ck/utility/random_gen.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..b7edf26507c62d5365ecab3fe1660c39a2cd672f
--- /dev/null
+++ b/include/ck/utility/random_gen.hpp
@@ -0,0 +1,53 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+namespace ck {
+
+// Pseudo random number generator
+// version for fp32
+template <typename T, uint32_t seed_t, std::enable_if_t<std::is_same<float, T>{}, bool> = false>
+__host__ __device__ uint32_t prand_generator(index_t id, T val, uint32_t seed = seed_t)
+{
+    uint32_t x         = *(reinterpret_cast<uint32_t*>(&val));
+    uint32_t drop_bits = uint32_t(x) & 0xFFFFu;
+    drop_bits ^= x >> 16;
+    drop_bits = ((drop_bits & 31) << 11) | (drop_bits >> 5);
+    drop_bits *= 0x7000149;
+    // NOTE: If id is in 64 bit, we are only using lower 32 bit.
+    //       So, it can have an effect of using same id for multiple elements when the id is very
+    //       large!
+    uint32_t rng = (drop_bits ^ 0x13371337 ^ (id * 229791) ^ seed);
+    return rng;
+}
+
+// version for fp16
+template <typename T, uint32_t seed_t, std::enable_if_t<std::is_same<half_t, T>{}, bool> = false>
+__host__ __device__ uint32_t prand_generator(index_t id, T val, uint32_t seed = seed_t)
+{
+    uint16_t x         = *(reinterpret_cast<uint16_t*>(&val));
+    uint32_t drop_bits = uint32_t(x) & 0xFFFFu;
+    drop_bits          = ((drop_bits & 31) << 11) | (drop_bits >> 5);
+    drop_bits *= 0x7000149;
+    // NOTE: If id is in 64 bit, we are only using lower 32 bit.
+    //       So, it can have an effect of using same id for multiple elements when the id is very
+    //       large!
+    uint32_t rng = (drop_bits ^ 0x13371337 ^ (id * 229791) ^ seed);
+    return rng;
+}
+
+// return 0 if data is not fp16 or fp32
+template <typename T,
+          uint32_t seed_t,
+          std::enable_if_t<!(std::is_same<float, T>{} || std::is_same<half_t, T>{}), bool> = false>
+__host__ __device__ uint32_t prand_generator(int id, T val, uint32_t seed = seed_t)
+{
+    std::ignore = id;
+    std::ignore = val;
+    std::ignore = seed;
+
+    return 0;
+}
+
+} // namespace ck
diff --git a/include/ck/utility/reduction_common.hpp b/include/ck/utility/reduction_common.hpp
index aceef7b296da7466c484aad8033fe1f949662fca..75fdd85825a71184e6f5afa4688a43c6c7a5518a 100644
--- a/include/ck/utility/reduction_common.hpp
+++ b/include/ck/utility/reduction_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -25,16 +25,4 @@ struct float_equal_zero
     };
 };
 
-template <index_t N>
-static constexpr __device__ index_t get_shift()
-{
-    return (get_shift<N / 2>() + 1);
-};
-
-template <>
-constexpr __device__ index_t get_shift<1>()
-{
-    return (0);
-}
-
 } // namespace ck
diff --git a/include/ck/utility/reduction_enums.hpp b/include/ck/utility/reduction_enums.hpp
index 67856331059cef628a62e4b2223f262b839bbf02..23b7149f8eb022a00d1aed81b4b69c315ca37ba0 100644
--- a/include/ck/utility/reduction_enums.hpp
+++ b/include/ck/utility/reduction_enums.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/reduction_functions_accumulate.hpp b/include/ck/utility/reduction_functions_accumulate.hpp
index 724e5599d6c878ba08bf14429c40c0958ab9f2a2..b9765ff0d2cb57bb2a1a0c4c7684a3cd2e0fa008 100644
--- a/include/ck/utility/reduction_functions_accumulate.hpp
+++ b/include/ck/utility/reduction_functions_accumulate.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/reduction_operator.hpp b/include/ck/utility/reduction_operator.hpp
index 25ae8fd34faf5452a29fb7c8271de01bae9badda..36c25203ea12b7a30b1991762c229da7cfcf417a 100644
--- a/include/ck/utility/reduction_operator.hpp
+++ b/include/ck/utility/reduction_operator.hpp
@@ -1,11 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck/ck.hpp"
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/type.hpp"
+#include "ck/utility/type_convert.hpp"
 
 namespace ck {
 
@@ -251,27 +252,27 @@ constexpr T GetIdentityValueForInMemoryDataOperation(InMemoryDataOperationEnum o
 };
 
 template <InMemoryDataOperationEnum Operation, typename DataType>
-struct InMemoryDataOperatonSupportedOnDataType
+struct InMemoryDataOperationSupportedOnDataType
 {
     static constexpr bool value = false;
 };
 
 template <typename DataType>
-struct InMemoryDataOperatonSupportedOnDataType<InMemoryDataOperationEnum::AtomicAdd, DataType>
+struct InMemoryDataOperationSupportedOnDataType<InMemoryDataOperationEnum::AtomicAdd, DataType>
 {
     static constexpr bool value =
         is_same<DataType, float>::value || is_same<DataType, double>::value;
 };
 
 template <typename DataType>
-struct InMemoryDataOperatonSupportedOnDataType<InMemoryDataOperationEnum::AtomicMax, DataType>
+struct InMemoryDataOperationSupportedOnDataType<InMemoryDataOperationEnum::AtomicMax, DataType>
 {
     static constexpr bool value =
         is_same<DataType, float>::value || is_same<DataType, double>::value;
 };
 
 template <typename DataType>
-struct InMemoryDataOperatonSupportedOnDataType<InMemoryDataOperationEnum::Set, DataType>
+struct InMemoryDataOperationSupportedOnDataType<InMemoryDataOperationEnum::Set, DataType>
 {
     static constexpr bool value =
         is_same<DataType, float>::value || is_same<DataType, double>::value ||
@@ -280,7 +281,7 @@ struct InMemoryDataOperatonSupportedOnDataType<InMemoryDataOperationEnum::Set, D
 };
 
 template <typename DataType>
-struct InMemoryDataOperatonSupportedOnDataType<InMemoryDataOperationEnum::Add, DataType>
+struct InMemoryDataOperationSupportedOnDataType<InMemoryDataOperationEnum::Add, DataType>
 {
     static constexpr bool value =
         is_same<DataType, float>::value || is_same<DataType, double>::value ||
diff --git a/include/ck/utility/sequence.hpp b/include/ck/utility/sequence.hpp
index 97b597221c2850d4b28b644266978fc56b295913..d6bfb2eba1cd6a6de2e334261379504d8448b3c7 100644
--- a/include/ck/utility/sequence.hpp
+++ b/include/ck/utility/sequence.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/sequence_helper.hpp b/include/ck/utility/sequence_helper.hpp
index db25c27e70c3653f94524367b8a6bce79480113e..8c493a28221eae4e095f051264f9323429984531 100644
--- a/include/ck/utility/sequence_helper.hpp
+++ b/include/ck/utility/sequence_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/span.hpp b/include/ck/utility/span.hpp
index 1e501214547cfbcec25921d6526e77563504835a..5e7567a847f3d18e424437005d876a409cd0d99d 100644
--- a/include/ck/utility/span.hpp
+++ b/include/ck/utility/span.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/static_buffer.hpp b/include/ck/utility/static_buffer.hpp
index dd25c96203288cfc9aa0959e9a7b1d83ef4abfc2..835f5657307a8e171b116c0bec088bb84b1b7b38 100644
--- a/include/ck/utility/static_buffer.hpp
+++ b/include/ck/utility/static_buffer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/statically_indexed_array.hpp b/include/ck/utility/statically_indexed_array.hpp
index 3438776f413cf010664cc8aa4a18c09bf161fae7..a2d70045a4b9b221651d958472aa6a2e721ccccc 100644
--- a/include/ck/utility/statically_indexed_array.hpp
+++ b/include/ck/utility/statically_indexed_array.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_STATICALLY_INDEXED_ARRAY_HPP
 #define CK_STATICALLY_INDEXED_ARRAY_HPP
diff --git a/include/ck/utility/statically_indexed_array_multi_index.hpp b/include/ck/utility/statically_indexed_array_multi_index.hpp
index 1079691f8c9d8f365536fc541d7eb70798248786..37c5e6a52944653a5df11c06a5e5dfad861dbeca 100644
--- a/include/ck/utility/statically_indexed_array_multi_index.hpp
+++ b/include/ck/utility/statically_indexed_array_multi_index.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_STATICALLY_INDEXED_ARRAY_MULTI_INDEX_HPP
 #define CK_STATICALLY_INDEXED_ARRAY_MULTI_INDEX_HPP
diff --git a/include/ck/utility/synchronization.hpp b/include/ck/utility/synchronization.hpp
index 9a463e56bb0f1f99838cd17de92e98d72d3089da..775e7ac3a392f5edd16054e63314bc80faca5d6e 100644
--- a/include/ck/utility/synchronization.hpp
+++ b/include/ck/utility/synchronization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -18,6 +18,7 @@ __device__ void block_sync_lds()
     __syncthreads();
 #endif
 }
+
 __device__ void s_nop()
 {
 #if 1
diff --git a/include/ck/utility/thread_group.hpp b/include/ck/utility/thread_group.hpp
index ae2db00ccf6b190a658c884d7c3c57f4fa326899..e2dad19a0b6750e47a682eefa08f719154ff8f3c 100644
--- a/include/ck/utility/thread_group.hpp
+++ b/include/ck/utility/thread_group.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/transpose_vectors.hpp b/include/ck/utility/transpose_vectors.hpp
index 2b0075d6005e3c5cf7d6772fda8132a6404878ed..6faf5c133ba2df0ed938aab915a272e3a7a590f3 100644
--- a/include/ck/utility/transpose_vectors.hpp
+++ b/include/ck/utility/transpose_vectors.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/tuple.hpp b/include/ck/utility/tuple.hpp
index d8664be550b79bc4de2edd224bc63350a0a1bde7..b616b3123f3b9ce3c92763cd36ecd39fc7d3e553 100644
--- a/include/ck/utility/tuple.hpp
+++ b/include/ck/utility/tuple.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/tuple_helper.hpp b/include/ck/utility/tuple_helper.hpp
index 6f5b142a5e7a765807bd9d7f556f7b8afc512d37..e39ae1c23d421e380b81b4bc7b487025cb6937be 100644
--- a/include/ck/utility/tuple_helper.hpp
+++ b/include/ck/utility/tuple_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/type.hpp b/include/ck/utility/type.hpp
index 90b9df2950b7d979f8cef386d155c72f6d7a39a5..9609afba43a78c05332d459b3a7cf756931a6e60 100644
--- a/include/ck/utility/type.hpp
+++ b/include/ck/utility/type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..8d6c703fb3a0fee3a6726a0adf27fd180fd0d4fe
--- /dev/null
+++ b/include/ck/utility/type_convert.hpp
@@ -0,0 +1,253 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/f8_utils.hpp"
+#include "ck/utility/random_gen.hpp"
+
+namespace ck {
+
+// Convert X to Y
+template <typename Y, typename X>
+__host__ __device__ constexpr Y type_convert(X x)
+{
+    static_assert(!std::is_reference_v<Y> && !std::is_reference_v<X>);
+
+    return static_cast<Y>(x);
+}
+
+// convert bfp16 to fp32
+template <>
+inline __host__ __device__ constexpr float type_convert<float, bhalf_t>(bhalf_t x)
+{
+    union
+    {
+        uint32_t int32;
+        float fp32;
+    } u = {uint32_t(x) << 16};
+
+    return u.fp32;
+}
+
+// convert fp32 to bfp16
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, float>(float x)
+{
+    union
+    {
+        float fp32;
+        uint32_t int32;
+    } u = {x};
+
+    return uint16_t(u.int32 >> 16);
+}
+
+// convert bfp16 to fp16 via fp32
+template <>
+inline __host__ __device__ constexpr half_t type_convert<half_t, bhalf_t>(bhalf_t x)
+{
+    float x_fp32 = type_convert<float>(x);
+
+    return static_cast<half_t>(x_fp32);
+}
+
+// convert fp16 to bfp16 via fp32
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, half_t>(half_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return type_convert<bhalf_t>(x_fp32);
+}
+
+// convert bfp16 to int32 via fp32
+template <>
+inline __host__ __device__ constexpr int32_t type_convert<int32_t, bhalf_t>(bhalf_t x)
+{
+    float x_fp32 = type_convert<float>(x);
+
+    return static_cast<int32_t>(x_fp32);
+}
+
+// convert int32 to bfp16 via fp32
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int32_t>(int32_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return type_convert<bhalf_t>(x_fp32);
+}
+
+// convert bfp16 to int8 via fp32
+template <>
+inline __host__ __device__ constexpr int8_t type_convert<int8_t, bhalf_t>(bhalf_t x)
+{
+    float x_fp32 = type_convert<float>(x);
+
+    return static_cast<int8_t>(x_fp32);
+}
+
+// convert int8 to bfp16 via fp32
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return type_convert<bhalf_t>(x_fp32);
+}
+
+// convert fp32 to fp8
+template <>
+inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+// convert fp8 to fp32
+template <>
+inline __host__ __device__ float type_convert<float, f8_t>(f8_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<float, negative_zero_nan>(x);
+}
+
+// convert fp16 to fp8
+template <>
+inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+// convert fp8 to fp16
+template <>
+inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<half_t, negative_zero_nan>(x);
+}
+
+template <>
+inline __host__ __device__ bhalf2_t type_convert<bhalf2_t, half2_t>(half2_t x)
+{
+    float y0{0}, y1{0};
+    bhalf2_t y{0};
+    asm volatile("\n \
+            v_cvt_f32_f16 %0, %1 \n \
+            "
+                 : "=v"(y0)
+                 : "v"(x));
+    asm volatile("\n \
+            v_cvt_f32_f16 %0, %1 src0_sel:WORD_1\n \
+            "
+                 : "=v"(y1)
+                 : "v"(x));
+    asm volatile("\n \
+            v_pack_b32_f16 %0, %1, %2 op_sel:[1, 1] \n \
+            "
+                 : "=v"(y)
+                 : "v"(y0), "v"(y1));
+    return y;
+}
+
+// Declare a template function for bf16 conversion using RTN
+template <typename Y, typename X>
+__host__ __device__ constexpr Y bf16_convert_rtn(X x);
+
+// Convert fp32 to bf16 with RTN if higher precision is needed
+template <>
+inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, float>(float x)
+{
+    union
+    {
+        float fp32;
+        uint32_t int32;
+    } u = {x};
+
+    // When the exponent bits are not all 1s, then the value is zero, normal,
+    // or subnormal. We round the bfloat16 mantissa up by adding 0x7FFF, plus
+    // 1 if the least significant bit of the bfloat16 mantissa is 1 (odd).
+    // This causes the bfloat16's mantissa to be incremented by 1 if the 16
+    // least significant bits of the float mantissa are greater than 0x8000,
+    // or if they are equal to 0x8000 and the least significant bit of the
+    // bfloat16 mantissa is 1 (odd). This causes it to be rounded to even when
+    // the lower 16 bits are exactly 0x8000. If the bfloat16 mantissa already
+    // has the value 0x7f, then incrementing it causes it to become 0x00 and
+    // the exponent is incremented by one, which is the next higher FP value
+    // to the unrounded bfloat16 value. When the bfloat16 value is subnormal
+    // with an exponent of 0x00 and a mantissa of 0x7f, it may be rounded up
+    // to a normal value with an exponent of 0x01 and a mantissa of 0x00.
+    // When the bfloat16 value has an exponent of 0xFE and a mantissa of 0x7F,
+    // incrementing it causes it to become an exponent of 0xFF and a mantissa
+    // of 0x00, which is Inf, the next higher value to the unrounded value.
+    bool flag0 = ~u.int32 & 0x7f800000;
+
+    // When all of the exponent bits are 1, the value is Inf or NaN.
+    // Inf is indicated by a zero mantissa. NaN is indicated by any nonzero
+    // mantissa bit. Quiet NaN is indicated by the most significant mantissa
+    // bit being 1. Signaling NaN is indicated by the most significant
+    // mantissa bit being 0 but some other bit(s) being 1. If any of the
+    // lower 16 bits of the mantissa are 1, we set the least significant bit
+    // of the bfloat16 mantissa, in order to preserve signaling NaN in case
+    // the bfloat16's mantissa bits are all 0.
+    bool flag1 = !flag0 && (u.int32 & 0xffff);
+
+    u.int32 += flag0 ? 0x7fff + ((u.int32 >> 16) & 1) : 0; // Round to nearest, round to even
+    u.int32 |= flag1 ? 0x10000 : 0x0;                      // Preserve signaling NaN
+
+    return uint16_t(u.int32 >> 16);
+}
+
+// convert fp16 to bfp16 via fp32 with RTN if higher precision is needed
+template <>
+inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, half_t>(half_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return bf16_convert_rtn<bhalf_t>(x_fp32);
+}
+
+// Declare a template function for fp8 conversion using SR
+template <typename Y, typename X>
+__host__ __device__ constexpr Y f8_convert_sr(X x);
+
+// convert fp32 to fp8 with stochastic rounding
+template <>
+inline __host__ __device__ f8_t f8_convert_sr<f8_t, float>(float x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    constexpr int seed               = 42;
+    // as thread id is not available on host, use 0 for prn generation
+    uint32_t rng = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
+    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+// convert fp16 to fp8 with stochastic rounding
+template <>
+inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    constexpr int seed               = 42;
+    // as thread id is not available on host, use 0 for prn generation
+    uint32_t rng = prand_generator<half_t, seed>(reinterpret_cast<uintptr_t>(&x), x);
+    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+} // namespace ck
diff --git a/include/ck/utility/workgroup_synchronization.hpp b/include/ck/utility/workgroup_synchronization.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..24858fdbdc65a978fec736f9c06ca31c5afd0a0a
--- /dev/null
+++ b/include/ck/utility/workgroup_synchronization.hpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+#include "ck/host_utility/hip_check_error.hpp"
+
+namespace ck {
+
+// Initialization flag of Barrier object, can be any value except for zero
+static constexpr int BarrierInitFlag = 0x7856;
+
+// 1) only the first thread-block in the synchronizaton group is supposed to call this function. It
+// is the responsibility of the user to ensure the two integer values in p_control_bits are zeros
+// before calling gms_init().
+// 2) Aftercalling gms_reset(), the two integer values in p_control_bits will be zeros, so no
+// repetitious initialization of p_control_bits buffer is required
+static __device__ void gms_init(int NumWarps, int* p_control_bits)
+{
+    union
+    {
+        int two32[2];
+        unsigned long one64;
+    } regs;
+
+    regs.two32[0] = BarrierInitFlag;
+    regs.two32[1] = NumWarps;
+
+    if(threadIdx.x == 0)
+        atomicCAS(reinterpret_cast<unsigned long*>(p_control_bits), 0, regs.one64);
+};
+
+// all the workgroups in the synchronization group is supposed to call this function
+static __device__ void gms_barrier(int* p_control_bits)
+{
+    constexpr int mask = warpSize - 1;
+
+    if((threadIdx.x & mask) == 0)
+    {
+        // ensure the barrier object is initialized
+        do
+        {
+            const int r0 = __atomic_load_n(&p_control_bits[0], __ATOMIC_RELAXED);
+
+            if(r0 == BarrierInitFlag)
+                break;
+
+        } while(true);
+
+        // go ahead toward the barrier line
+        atomicSub(&p_control_bits[1], 1);
+
+        // wait until all warps have arrived
+        do
+        {
+            const int r1 = __atomic_load_n(&p_control_bits[1], __ATOMIC_RELAXED);
+
+            if(r1 == 0)
+                break;
+
+        } while(true);
+    };
+};
+
+// 1) Only the first thread-block in the synchronizaton group is supposed to call this function.
+// 2) Aftercalling gms_reset(), the two integer values in p_control_bits will be zeros, so no
+// repetitious initialization of p_control_bits buffer is required
+static __device__ void gms_reset(int* p_control_bits)
+{
+    // reset the barrier object
+    if(threadIdx.x == 0)
+        (void)atomicCAS(&p_control_bits[0], BarrierInitFlag, 0);
+};
+
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp
index 46a1fa559a1a3d164acc99e536f7e26a83b97ed4..a1b1e0d91b472e5531097b13d57d6efe5b9591a7 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp
index 0b621e88a0c6e890bf1efb026615fcc63fadf6c9..a2eabdf5c1c92375503fc8cc961a2cde8410204e 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp
index dd0db316804fd191ee73d9952f6df6779420910d..20c1fcd7367838022d7a4e2776f436ce796460fe 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp
index 463c655ac1d1414f80f9d8b7fcfc696be6326018..7d652fe4c41d044747c2c40a3d6e7383d4f1ff2e 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp
index b0149d88fdb062806d3b34b400f9b6d3389419f9..24f754e5987adfa82f4b0cc0e41dc85431f94e12 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..92a8c82a6e81c85ebe79aa8941f51d0a53e4734d
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+// hardcoded for NumDimM == NumDimN == NumDimK == 2
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
+struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
+{
+    // Argument
+    struct Argument : public ck::tensor_operation::device::BaseArgument
+    {
+        Argument(const Tensor<ADataType>& a_ms_ks,
+                 const Tensor<BDataType>& b_ns_ks,
+                 Tensor<CDataType>& c_ms_ns,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op)
+            : a_ms_ks_{a_ms_ks},
+              b_ns_ks_{b_ns_ks},
+              c_ms_ns_{c_ms_ns},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op}
+        {
+        }
+
+        const Tensor<ADataType>& a_ms_ks_;
+        const Tensor<BDataType>& b_ns_ks_;
+        Tensor<CDataType>& c_ms_ns_;
+
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+    };
+
+    // Invoker
+    struct Invoker : public ck::tensor_operation::device::BaseInvoker
+    {
+        using Argument = ReferenceContraction_M2_N2_K2::Argument;
+
+        float Run(const Argument& arg)
+        {
+            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
+                const ck::index_t K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
+                const ck::index_t K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
+
+                AccDataType v_acc = 0;
+
+                for(ck::index_t k0 = 0; k0 < K0; ++k0)
+                {
+                    for(ck::index_t k1 = 0; k1 < K1; ++k1)
+                    {
+                        AccDataType v_a;
+                        AccDataType v_b;
+
+                        arg.a_element_op_(
+                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
+                        arg.b_element_op_(
+                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
+
+                        v_acc += v_a * v_b;
+                    }
+                }
+
+                arg.c_ms_ns_(m0, m1, n0, n1) = v_acc;
+            };
+
+            make_ParallelTensorFunctor(f_ms_ns,
+                                       arg.c_ms_ns_.mDesc.GetLengths()[0],
+                                       arg.c_ms_ns_.mDesc.GetLengths()[1],
+                                       arg.c_ms_ns_.mDesc.GetLengths()[2],
+                                       arg.c_ms_ns_.mDesc.GetLengths()[3])(
+                std::thread::hardware_concurrency());
+
+            return 0;
+        }
+
+        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
+    {
+        return true;
+    }
+
+    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
+                             const Tensor<BDataType>& b_ns_ks,
+                             Tensor<CDataType>& c_ms_ns,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op)
+    {
+        return Argument{a_ms_ks, b_ns_ks, c_ms_ns, a_element_op, b_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceContraction_M2_N2_K2"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
index 225f7b7e36f3a5a0d70058ed7089fc693a9e0493..449734f43478f0b36215d9d876c4a7537675a13f 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
index 7d62158f00c0aa28fdfe90d4a9df537367b22717..ec5df238ab2c2a292d8d44951f67e2d8d2383d1a 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
index b8d47d218b95e3ff2b38c2fc965b6bdc6a2afc68..8f4182a2318017dc7239dcc32e34af5287e97b4d 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp
index be22003fd90db20eaea80045bd34f908ed8ebb61..71c84a1f5cd43811105a4630e31ade8ae0b3a075 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp
index f949f27fde973cc44eec3162c7625980a9818116..0b90b4b50eec00cbd4d807fe2f43b9280cd2bdf2 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
index 6728bb1f471b4aad3f7ed0a2878fc7a81e1b569d..9b797be92549703fff3397fe9f099ad3e76c8389 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
@@ -1,11 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include <iostream>
 #include <sstream>
 
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/device/device_base.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 
@@ -66,8 +67,26 @@ struct ReferenceGemm : public device::BaseOperator
                     ADataType v_a;
                     BDataType v_b;
 
-                    arg.a_element_op_(v_a, arg.a_m_k_(m, k));
-                    arg.b_element_op_(v_b, arg.b_k_n_(k, n));
+                    // use PassThrough instead of ConvertBF16RTN for reference calculation
+                    if constexpr(is_same_v<AElementwiseOperation,
+                                           ck::tensor_operation::element_wise::ConvertBF16RTN>)
+                    {
+                        ck::tensor_operation::element_wise::PassThrough{}(v_a, arg.a_m_k_(m, k));
+                    }
+                    else
+                    {
+                        arg.a_element_op_(v_a, arg.a_m_k_(m, k));
+                    }
+                    // same for B matrix
+                    if constexpr(is_same_v<BElementwiseOperation,
+                                           ck::tensor_operation::element_wise::ConvertBF16RTN>)
+                    {
+                        ck::tensor_operation::element_wise::PassThrough{}(v_b, arg.b_k_n_(k, n));
+                    }
+                    else
+                    {
+                        arg.b_element_op_(v_b, arg.b_k_n_(k, n));
+                    }
 
                     v_acc +=
                         ck::type_convert<AccDataType>(v_a) * ck::type_convert<AccDataType>(v_b);
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_2d.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_2d.hpp
deleted file mode 100644
index c77d22f4cd1c34d3fa81a973f115fc37b7d03cf0..0000000000000000000000000000000000000000
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_2d.hpp
+++ /dev/null
@@ -1,136 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace host {
-
-template <typename ADataType,
-          typename BDataType,
-          typename C0DataType,
-          typename CDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation>
-struct ReferenceGemmBias2D : public device::BaseOperator
-{
-    // Argument
-    struct Argument : public device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_m_k,
-                 const Tensor<BDataType>& b_k_n,
-                 const Tensor<C0DataType>& c0_m_n,
-                 Tensor<CDataType>& c_m_n,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : a_m_k_{a_m_k},
-              b_k_n_{b_k_n},
-              c0_m_n_{c0_m_n},
-              c_m_n_{c_m_n},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_m_k_;
-        const Tensor<BDataType>& b_k_n_;
-        const Tensor<CDataType>& c0_m_n_;
-        Tensor<CDataType>& c_m_n_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public device::BaseInvoker
-    {
-        using Argument = ReferenceGemmBias2D::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_mk_kn_mn = [&](auto m, auto n) {
-                const int K = arg.a_m_k_.mDesc.GetLengths()[1];
-
-                AccDataType a   = 0;
-                AccDataType b   = 0;
-                AccDataType acc = 0;
-
-                for(int k = 0; k < K; ++k)
-                {
-                    arg.a_element_op_(a, ck::type_convert<AccDataType>(arg.a_m_k_(m, k)));
-                    arg.b_element_op_(b, ck::type_convert<AccDataType>(arg.b_k_n_(k, n)));
-                    acc += a * b;
-                }
-
-                CDataType cast_acc = static_cast<CDataType>(acc);
-                arg.c_element_op_(arg.c_m_n_(m, n), cast_acc, arg.c0_m_n_(m, n));
-            };
-
-            make_ParallelTensorFunctor(
-                f_mk_kn_mn, arg.c_m_n_.mDesc.GetLengths()[0], arg.c_m_n_.mDesc.GetLengths()[1])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_m_k,
-                             const Tensor<BDataType>& b_k_n,
-                             const Tensor<C0DataType>& c0_m_n,
-                             Tensor<CDataType>& c_m_n,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
-    {
-        return Argument{a_m_k, b_k_n, c0_m_n, c_m_n, a_element_op, b_element_op, c_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceGemmBias2D"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace host
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp
deleted file mode 100644
index 7dfc3c1ed4b868756277f43d341cca5a1796662e..0000000000000000000000000000000000000000
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp
+++ /dev/null
@@ -1,140 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace host {
-
-template <typename ADataType,
-          typename BDataType,
-          typename CDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation>
-struct ReferenceGemmBiasActivation : public device::BaseOperator
-{
-    // Argument
-    struct Argument : public device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_m_k,
-                 const Tensor<BDataType>& b_k_n,
-                 Tensor<CDataType>& c_m_n,
-                 const Tensor<CDataType>& c0_n,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : a_m_k_{a_m_k},
-              b_k_n_{b_k_n},
-              c_m_n_{c_m_n},
-              c0_n_{c0_n},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_m_k_;
-        const Tensor<BDataType>& b_k_n_;
-        Tensor<CDataType>& c_m_n_;
-        const Tensor<CDataType>& c0_n_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public device::BaseInvoker
-    {
-        using Argument = ReferenceGemmBiasActivation::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_mk_kn_mn = [&](auto m, auto n) {
-                const int K = arg.a_m_k_.mDesc.GetLengths()[1];
-
-                float v_acc = 0;
-
-                for(int k = 0; k < K; ++k)
-                {
-                    float v_a;
-                    float v_b;
-
-                    arg.a_element_op_(v_a, static_cast<const float>(arg.a_m_k_(m, k)));
-                    arg.b_element_op_(v_b, static_cast<const float>(arg.b_k_n_(k, n)));
-
-                    v_acc += v_a * v_b;
-                }
-
-                float v_c;
-
-                arg.c_element_op_(v_c, v_acc, static_cast<float>(arg.c0_n_(n)));
-
-                arg.c_m_n_(m, n) = v_c;
-            };
-
-            make_ParallelTensorFunctor(
-                f_mk_kn_mn, arg.c_m_n_.mDesc.GetLengths()[0], arg.c_m_n_.mDesc.GetLengths()[1])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_m_k,
-                             const Tensor<BDataType>& b_k_n,
-                             Tensor<CDataType>& c_m_n,
-                             const Tensor<CDataType>& c0_n,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
-    {
-        return Argument{a_m_k, b_k_n, c_m_n, c0_n, a_element_op, b_element_op, c_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceGemmBiasActivation"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace host
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp
deleted file mode 100644
index 99102a40d4e9a4c6360b3e6ecf9c8a345f5b38a9..0000000000000000000000000000000000000000
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp
+++ /dev/null
@@ -1,148 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace host {
-
-template <typename ADataType,
-          typename BDataType,
-          typename CDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation>
-struct ReferenceGemmBiasActivationAdd : public device::BaseOperator
-{
-    // Argument
-    struct Argument : public device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_m_k,
-                 const Tensor<BDataType>& b_k_n,
-                 Tensor<CDataType>& c_m_n,
-                 const Tensor<CDataType>& c0_n,
-                 const Tensor<CDataType>& c1_m_n,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : a_m_k_{a_m_k},
-              b_k_n_{b_k_n},
-              c_m_n_{c_m_n},
-              c0_n_{c0_n},
-              c1_m_n_{c1_m_n},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_m_k_;
-        const Tensor<BDataType>& b_k_n_;
-        Tensor<CDataType>& c_m_n_;
-        const Tensor<CDataType>& c0_n_;
-        const Tensor<CDataType>& c1_m_n_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public device::BaseInvoker
-    {
-        using Argument = ReferenceGemmBiasActivationAdd::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_mk_kn_mn = [&](auto m, auto n) {
-                const int K = arg.a_m_k_.mDesc.GetLengths()[1];
-
-                float v_acc = 0;
-
-                for(int k = 0; k < K; ++k)
-                {
-                    float v_a;
-                    float v_b;
-
-                    arg.a_element_op_(v_a, static_cast<const float>(arg.a_m_k_(m, k)));
-                    arg.b_element_op_(v_b, static_cast<const float>(arg.b_k_n_(k, n)));
-
-                    v_acc += v_a * v_b;
-                }
-
-                float v_c;
-
-                arg.c_element_op_(v_c,
-                                  v_acc,
-                                  static_cast<float>(arg.c0_n_(n)),
-                                  static_cast<float>(arg.c1_m_n_(m, n)));
-
-                arg.c_m_n_(m, n) = v_c;
-            };
-
-            make_ParallelTensorFunctor(
-                f_mk_kn_mn, arg.c_m_n_.mDesc.GetLengths()[0], arg.c_m_n_.mDesc.GetLengths()[1])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_m_k,
-                             const Tensor<BDataType>& b_k_n,
-                             Tensor<CDataType>& c_m_n,
-                             const Tensor<CDataType>& c0_n,
-                             const Tensor<CDataType>& c1_m_n,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
-    {
-        return Argument{
-            a_m_k, b_k_n, c_m_n, c0_n, c1_m_n, a_element_op, b_element_op, c_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceGemmBiasActivationAdd"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace host
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp
index 28132aa1ebd78178774caba73b91ba3c947b66be..ce2a83da6163fb6b19e30121a48734096857c508 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
index fedd4dce62cdcd31c2c582ab034291e9e098bc3c..6a48528c543b3ad1bcf94f92effcacbce40a8151 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
index 680d94f7d1dcc3c474ba665887a132693e323c0f..9994a2f9f7c862a1bf870103f999634e17682ecf 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -90,10 +90,13 @@ struct ReferenceLayernorm : public device::BaseOperator
 
             for(int m = 0; m < M; ++m)
             {
+                AccDataType divisor =
+                    static_cast<AccDataType>(1) / ck::math::sqrt(var(m) + arg.epsilon_);
+
                 for(int n = 0; n < N; ++n)
                 {
                     auto x_val = ck::type_convert<AccDataType>(arg.x_m_n_(m, n));
-                    auto y_val = (x_val - mean(m)) / sqrt(var(m) + arg.epsilon_);
+                    auto y_val = (x_val - mean(m)) * divisor;
                     y_val      = (y_val * arg.gamma_n_(n)) + arg.beta_n_(n);
                     arg.acc_elementwise_op_(y_val, y_val);
                     arg.y_m_n_(m, n) = ck::type_convert<YDataType>(y_val);
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..3f1fc6165cd00daf9f2c8ae731dfc62a71d2ccd6
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp
@@ -0,0 +1,103 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+using namespace std;
+
+template <typename DOutDataType,
+          typename IndexDataType,
+          typename ConputeDataType,
+          typename DInDataType,
+          typename ElementwiseOperation>
+struct ReferenceMaxPoolBwd : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        Argument(const Tensor<DOutDataType>& dout,
+                 const Tensor<IndexDataType>& indices,
+                 Tensor<DInDataType>& din,
+                 ElementwiseOperation elementwise_op)
+            : dout_(dout), indices_(indices), din_(din), elementwise_op_(elementwise_op)
+        {
+        }
+
+        const Tensor<DOutDataType>& dout_;
+        const Tensor<IndexDataType>& indices_;
+        Tensor<DInDataType>& din_;
+        ElementwiseOperation elementwise_op_;
+    };
+
+    // Invoker
+    struct Invoker : public device::BaseInvoker
+    {
+        float Run(const Argument& arg)
+        {
+            int din_length  = arg.din_.GetElementSpaceSize();
+            int dout_length = arg.dout_.GetElementSpaceSize();
+            std::vector<ConputeDataType> buf(din_length, 0);
+
+            for(int i = 0; i < dout_length; ++i)
+            {
+                int index = arg.indices_.mData[i];
+                if(index >= 0 && index < din_length)
+                    buf[index] += ck::type_convert<ConputeDataType>(arg.dout_.mData[i]);
+            }
+
+            for(int i = 0; i < din_length; ++i)
+                arg.din_.mData[i] = ck::type_convert<DInDataType>(buf[i]);
+            return 0;
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
+
+    static auto MakeArgument(const Tensor<DOutDataType>& dout,
+                             const Tensor<IndexDataType>& indices,
+                             Tensor<DInDataType>& din,
+                             ElementwiseOperation elementwise_op)
+    {
+        return Argument{dout, indices, din, elementwise_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceMaxPoolBwd"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..696fb5eaf1413755fd3990f5db4e9b34f72fe654
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
@@ -0,0 +1,345 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+#include <vector>
+#include <algorithm>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+#include "ck/utility/reduction_functions_accumulate.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+template <index_t InOutRank,
+          index_t WindowRank,
+          typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+struct ReferencePoolingFwd : public device::BaseOperator
+{
+    using ReduceOperation = typename ck::reduce_binary_operator<ReduceOpId>::opType;
+
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        Argument(const Tensor<InDataType>& in,
+                 Tensor<OutDataType>& out,
+                 Tensor<IndexDataType>& out_indices,
+                 const std::vector<ck::index_t>& window_spatial_lengths,
+                 const std::vector<ck::index_t>& window_strides,
+                 const std::vector<ck::index_t>& in_left_pads,
+                 const std::vector<ck::index_t>& /*in_right_pads*/)
+            : in_(in),
+              out_(out),
+              out_indices_(out_indices),
+              window_spatial_lengths_(window_spatial_lengths),
+              window_strides_(window_strides),
+              in_left_pads_(in_left_pads),
+              reduceLength_(1)
+        {
+            static_for<0, WindowRank, 1>{}(
+                [&](auto I) { reduceLength_ *= window_spatial_lengths[I]; });
+        }
+
+        const Tensor<InDataType>& in_;
+        Tensor<OutDataType>& out_;
+        Tensor<IndexDataType>& out_indices_;
+        const std::vector<ck::index_t>& window_spatial_lengths_;
+        const std::vector<ck::index_t>& window_strides_;
+        const std::vector<ck::index_t>& in_left_pads_;
+        int reduceLength_;
+    };
+
+    // Invoker
+    struct Invoker : public device::BaseInvoker
+    {
+        float RunPooling3dFwd(const Argument& arg)
+        {
+
+            auto elementwise_ops =
+                ck::reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
+                    arg.reduceLength_);
+
+            auto in_elementwise_op  = std::get<0>(elementwise_ops);
+            auto acc_elementwise_op = std::get<1>(elementwise_ops);
+
+            if constexpr(!OutputIndex)
+            {
+                using Accumulation = ck::detail::
+                    AccumulateWithNanCheck<PropagateNan, ReduceOperation, ComputeDataType>;
+
+                auto f_ncdhw = [&](auto n, auto c, auto do_, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+
+                    for(ck::index_t z = 0; z < arg.window_spatial_lengths_[0]; ++z)
+                    {
+                        ck::index_t di = do_ * arg.window_strides_[0] + z - arg.in_left_pads_[0];
+                        for(ck::index_t y = 0; y < arg.window_spatial_lengths_[1]; ++y)
+                        {
+                            ck::index_t hi = ho * arg.window_strides_[1] + y - arg.in_left_pads_[1];
+                            for(ck::index_t x = 0; x < arg.window_spatial_lengths_[2]; ++x)
+                            {
+                                ck::index_t wi =
+                                    wo * arg.window_strides_[2] + x - arg.in_left_pads_[2];
+                                if(di >= 0 &&
+                                   di < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                                   hi >= 0 &&
+                                   hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]) &&
+                                   wi >= 0 &&
+                                   wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[4]))
+                                {
+                                    ComputeDataType currVal = ck::type_convert<ComputeDataType>(
+                                        arg.in_(n, c, di, hi, wi));
+
+                                    in_elementwise_op(currVal, currVal);
+
+                                    Accumulation::Calculate(accuVal, currVal);
+                                }
+                            }
+                        }
+                    }
+                    acc_elementwise_op(accuVal, accuVal);
+
+                    arg.out_(n, c, do_, ho, wo) = ck::type_convert<OutDataType>(accuVal);
+                };
+
+                make_ParallelTensorFunctor(f_ncdhw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3],
+                                           arg.out_.mDesc.GetLengths()[4])(
+                    std::thread::hardware_concurrency());
+            }
+            else
+            {
+                using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
+                                                                                ReduceOperation,
+                                                                                ComputeDataType,
+                                                                                IndexDataType>;
+
+                auto f_ncdhw = [&](auto n, auto c, auto do_, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+                    IndexDataType accuIndex = 0;
+
+                    for(ck::index_t z = 0; z < arg.window_spatial_lengths_[0]; ++z)
+                    {
+                        ck::index_t di = do_ * arg.window_strides_[0] + z - arg.in_left_pads_[0];
+                        for(ck::index_t y = 0; y < arg.window_spatial_lengths_[1]; ++y)
+                        {
+                            ck::index_t hi = ho * arg.window_strides_[1] + y - arg.in_left_pads_[1];
+                            for(ck::index_t x = 0; x < arg.window_spatial_lengths_[2]; ++x)
+                            {
+                                ck::index_t wi =
+                                    wo * arg.window_strides_[2] + x - arg.in_left_pads_[2];
+                                if(di >= 0 &&
+                                   di < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                                   hi >= 0 &&
+                                   hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]) &&
+                                   wi >= 0 &&
+                                   wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[4]))
+                                {
+                                    ComputeDataType currVal = ck::type_convert<ComputeDataType>(
+                                        arg.in_(n, c, di, hi, wi));
+                                    IndexDataType currIndex =
+                                        arg.in_.GetOffsetFromMultiIndex(n, c, di, hi, wi);
+
+                                    in_elementwise_op(currVal, currVal);
+
+                                    Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
+                                }
+                            }
+                        }
+                    }
+
+                    acc_elementwise_op(accuVal, accuVal);
+
+                    arg.out_(n, c, do_, ho, wo)         = ck::type_convert<OutDataType>(accuVal);
+                    arg.out_indices_(n, c, do_, ho, wo) = accuIndex;
+                };
+
+                make_ParallelTensorFunctor(f_ncdhw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3],
+                                           arg.out_.mDesc.GetLengths()[4])(
+                    std::thread::hardware_concurrency());
+            };
+
+            return 0;
+        }
+
+        float RunPooling2dFwd(const Argument& arg)
+        {
+
+            auto elementwise_ops =
+                ck::reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
+                    arg.reduceLength_);
+
+            auto in_elementwise_op  = std::get<0>(elementwise_ops);
+            auto acc_elementwise_op = std::get<1>(elementwise_ops);
+
+            if constexpr(!OutputIndex)
+            {
+                using Accumulation = ck::detail::
+                    AccumulateWithNanCheck<PropagateNan, ReduceOperation, ComputeDataType>;
+
+                auto f_nchw = [&](auto n, auto c, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+
+                    for(ck::index_t y = 0; y < arg.window_spatial_lengths_[0]; ++y)
+                    {
+                        ck::index_t hi = ho * arg.window_strides_[0] + y - arg.in_left_pads_[0];
+                        for(ck::index_t x = 0; x < arg.window_spatial_lengths_[1]; ++x)
+                        {
+                            ck::index_t wi = wo * arg.window_strides_[1] + x - arg.in_left_pads_[1];
+                            if(hi >= 0 &&
+                               hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                               wi >= 0 &&
+                               wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]))
+                            {
+                                ComputeDataType currVal =
+                                    ck::type_convert<ComputeDataType>(arg.in_(n, c, hi, wi));
+
+                                in_elementwise_op(currVal, currVal);
+
+                                Accumulation::Calculate(accuVal, currVal);
+                            }
+                        }
+                    }
+
+                    acc_elementwise_op(accuVal, accuVal);
+                    arg.out_(n, c, ho, wo) = ck::type_convert<OutDataType>(accuVal);
+                };
+
+                make_ParallelTensorFunctor(f_nchw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3])(
+                    std::thread::hardware_concurrency());
+            }
+            else
+            {
+                using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
+                                                                                ReduceOperation,
+                                                                                ComputeDataType,
+                                                                                IndexDataType>;
+
+                auto f_nchw = [&](auto n, auto c, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+                    IndexDataType accuIndex = 0;
+
+                    for(ck::index_t y = 0; y < arg.window_spatial_lengths_[0]; ++y)
+                    {
+                        ck::index_t hi = ho * arg.window_strides_[0] + y - arg.in_left_pads_[0];
+                        for(ck::index_t x = 0; x < arg.window_spatial_lengths_[1]; ++x)
+                        {
+                            ck::index_t wi = wo * arg.window_strides_[1] + x - arg.in_left_pads_[1];
+                            if(hi >= 0 &&
+                               hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                               wi >= 0 &&
+                               wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]))
+                            {
+                                ComputeDataType currVal =
+                                    ck::type_convert<ComputeDataType>(arg.in_(n, c, hi, wi));
+
+                                IndexDataType currIndex =
+                                    arg.in_.GetOffsetFromMultiIndex(n, c, hi, wi);
+
+                                in_elementwise_op(currVal, currVal);
+
+                                Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
+                            }
+                        }
+                    }
+
+                    acc_elementwise_op(accuVal, accuVal);
+                    arg.out_(n, c, ho, wo)         = ck::type_convert<OutDataType>(accuVal);
+                    arg.out_indices_(n, c, ho, wo) = accuIndex;
+                };
+
+                make_ParallelTensorFunctor(f_nchw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3])(
+                    std::thread::hardware_concurrency());
+            };
+
+            return 0;
+        }
+
+        float Run(const Argument& arg)
+        {
+            // TODO - support generic pooling
+            if constexpr(InOutRank == 5 && WindowRank == 3)
+                return RunPooling3dFwd(arg);
+            else if constexpr(InOutRank == 4 && WindowRank == 2)
+                return RunPooling2dFwd(arg);
+            else
+                throw std::runtime_error("Only support pooling3d or pooling2d so far");
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
+
+    static auto MakeArgument(const Tensor<InDataType>& in,
+                             Tensor<OutDataType>& out,
+                             Tensor<IndexDataType>& out_indices,
+                             const std::vector<ck::index_t>& window_spatial_lengths,
+                             const std::vector<ck::index_t>& window_strides,
+                             const std::vector<ck::index_t>& in_left_pads,
+                             const std::vector<ck::index_t>& in_right_pads)
+    {
+        return Argument{in,
+                        out,
+                        out_indices,
+                        window_spatial_lengths,
+                        window_strides,
+                        in_left_pads,
+                        in_right_pads};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferencePoolingFwd"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_reduce.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_reduce.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..944f34007ef0e030e2017688477c27602df15391
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_reduce.hpp
@@ -0,0 +1,435 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <array>
+#include <algorithm>
+#include <thread>
+
+#include "ck/ck.hpp"
+#include "ck/utility/ignore.hpp"
+#include "ck/utility/reduction_common.hpp"
+#include "ck/utility/reduction_functions_accumulate.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/tensor_operation/gpu/device/device_reduce.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
+          index_t NumReduceDim,
+          typename ReduceOperation,
+          typename InElementwiseOperation,
+          typename AccElementwiseOperation,
+          bool PropagateNan,
+          bool OutputIndex>
+struct ReferenceReduce : public device::DeviceReduce<InDataType,
+                                                     AccDataType,
+                                                     OutDataType,
+                                                     Rank,
+                                                     NumReduceDim,
+                                                     ReduceOperation,
+                                                     InElementwiseOperation,
+                                                     AccElementwiseOperation,
+                                                     PropagateNan,
+                                                     OutputIndex>
+{
+    using IndexDataType = int32_t;
+
+    static constexpr int NumInvariantDim = Rank - NumReduceDim;
+
+    static constexpr index_t NumSrcDim = Rank;
+    static constexpr index_t NumDstDim = (NumInvariantDim == 0) ? 1 : NumInvariantDim;
+    static constexpr bool reduceAllDim = (NumInvariantDim == 0);
+
+    struct Argument : public device::BaseArgument
+    {
+        Argument(const std::array<index_t, Rank> inLengths,
+                 const std::array<index_t, Rank> inStrides,
+                 const std::array<index_t, NumDstDim> outLengths,
+                 const std::array<index_t, NumDstDim> outStrides,
+                 const std::array<int, NumReduceDim> reduceDims,
+                 double alpha,
+                 double beta,
+                 const InDataType* in_host,
+                 OutDataType* out_host,
+                 IndexDataType* out_index_host,
+                 const InElementwiseOperation in_elementwise_op,
+                 const AccElementwiseOperation acc_elementwise_op)
+            : reduceDims_(reduceDims),
+              outLengths_(outLengths),
+              outStrides_(outStrides),
+              in_host_(in_host),
+              out_host_(out_host),
+              out_index_host_(out_index_host),
+              in_elementwise_op_(in_elementwise_op),
+              acc_elementwise_op_(acc_elementwise_op)
+        {
+            using ck::host_common::get_index_set;
+
+            if(std::any_of(
+                   reduceDims.begin(), reduceDims.end(), [](int d) { return d < 0 || d >= Rank; }))
+                throw std::runtime_error("Invalid reduce dimensions!");
+
+            if constexpr(NumInvariantDim > 0)
+            {
+                // get invariant_dims[] and invariant_lengths[]
+                for(int dim = 0, i = 0; dim < Rank; dim++)
+                    if(std::none_of(
+                           reduceDims.begin(), reduceDims.end(), [&](int d) { return d == dim; }))
+                    {
+                        invariantDims_[i]     = dim;
+                        invariant_lengths_[i] = inLengths[dim];
+                        i++;
+                    };
+            };
+
+            // get reduce_lengths_[]
+            for(int j = 0, i = 0; j < NumReduceDim; j++)
+            {
+                int dim              = reduceDims[j];
+                reduce_lengths_[i++] = inLengths[dim];
+            };
+
+            if constexpr(NumInvariantDim > 0)
+            {
+                // check invariant_lengths_ and outLengths
+                for(int i = 0; i < NumInvariantDim; i++)
+                    if(invariant_lengths_[i] != outLengths_[i])
+                        throw std::runtime_error("Invalid lengths parameters!");
+            }
+
+            if constexpr(NumInvariantDim > 0)
+            {
+                for(int j = 0, i = 0; j < NumInvariantDim; j++)
+                {
+                    int dim                  = invariantDims_[j];
+                    in_invariant_strides_[i] = inStrides[dim];
+                    i++;
+                };
+            };
+
+            for(int j = 0, i = 0; j < NumReduceDim; j++)
+            {
+                int dim               = reduceDims_[j];
+                in_reduce_strides_[i] = inStrides[dim];
+                i++;
+            };
+
+            if constexpr(NumInvariantDim > 0)
+                invariant_index_set_ = get_index_set<NumInvariantDim>(invariant_lengths_);
+
+            reduce_index_set_ = get_index_set<NumReduceDim>(reduce_lengths_);
+
+            alpha_ = type_convert<AccDataType>(alpha);
+            beta_  = type_convert<AccDataType>(beta);
+        };
+
+        const std::array<int, NumReduceDim> reduceDims_;
+        std::array<int, NumInvariantDim> invariantDims_;
+        std::array<index_t, NumInvariantDim> invariant_lengths_;
+        std::array<index_t, NumReduceDim> reduce_lengths_;
+
+        const std::array<index_t, NumDstDim> outLengths_;
+        const std::array<index_t, NumDstDim> outStrides_;
+
+        std::array<index_t, NumInvariantDim> in_invariant_strides_;
+        std::array<index_t, NumReduceDim> in_reduce_strides_;
+
+        const InDataType* in_host_;
+        OutDataType* out_host_;
+        IndexDataType* out_index_host_;
+        const InElementwiseOperation in_elementwise_op_;
+        const AccElementwiseOperation acc_elementwise_op_;
+
+        AccDataType alpha_;
+        AccDataType beta_;
+
+        std::vector<std::array<index_t, NumInvariantDim>> invariant_index_set_;
+        std::vector<std::array<index_t, NumReduceDim>> reduce_index_set_;
+    };
+
+    struct Invoker : public device::BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            ignore = stream_config;
+
+            using ck::float_equal_one;
+            using ck::float_equal_zero;
+            using ck::type_convert;
+            using ck::host_common::get_index_set;
+            using ck::host_common::get_offset_from_index;
+
+            if constexpr(OutputIndex)
+            {
+                using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
+                                                                                ReduceOperation,
+                                                                                AccDataType,
+                                                                                IndexDataType>;
+
+                if constexpr(NumInvariantDim == 0)
+                {
+                    AccDataType accuVal = ReduceOperation::template GetIdentityValue<AccDataType>();
+                    IndexDataType accuIndex = 0;
+
+                    for(std::size_t i = 0; i < arg.reduce_index_set_.size(); i++)
+                    {
+                        auto in_offset = get_offset_from_index<NumReduceDim>(
+                            arg.in_reduce_strides_, arg.reduce_index_set_[i]);
+
+                        auto currVal = type_convert<AccDataType>(arg.in_host_[in_offset]);
+
+                        arg.in_elementwise_op_(currVal, currVal);
+
+                        auto currIndex = static_cast<IndexDataType>(i);
+
+                        Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
+                    };
+
+                    arg.acc_elementwise_op_(accuVal, accuVal);
+
+                    if(!float_equal_one{}(arg.alpha_))
+                        accuVal *= type_convert<AccDataType>(arg.alpha_);
+
+                    if(!float_equal_zero{}(arg.beta_))
+                        accuVal += type_convert<AccDataType>(arg.out_host_[0]) *
+                                   type_convert<AccDataType>(arg.beta_);
+
+                    arg.out_host_[0]       = type_convert<OutDataType>(accuVal);
+                    arg.out_index_host_[0] = accuIndex;
+                }
+                else
+                {
+                    auto thread_reduce_func = [&](auto invariant_index) {
+                        AccDataType accuVal =
+                            ReduceOperation::template GetIdentityValue<AccDataType>();
+                        IndexDataType accuIndex = 0;
+
+                        auto in_invariant_offset = get_offset_from_index<NumInvariantDim>(
+                            arg.in_invariant_strides_, invariant_index);
+
+                        for(std::size_t i = 0; i < arg.reduce_index_set_.size(); i++)
+                        {
+                            auto in_reduce_offset = get_offset_from_index<NumReduceDim>(
+                                arg.in_reduce_strides_, arg.reduce_index_set_[i]);
+
+                            auto currVal = type_convert<AccDataType>(
+                                arg.in_host_[in_invariant_offset + in_reduce_offset]);
+
+                            arg.in_elementwise_op_(currVal, currVal);
+
+                            auto currIndex = static_cast<IndexDataType>(i);
+
+                            Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
+                        };
+
+                        arg.acc_elementwise_op_(accuVal, accuVal);
+
+                        if(!float_equal_one{}(arg.alpha_))
+                            accuVal *= type_convert<AccDataType>(arg.alpha_);
+
+                        auto dst_offset = get_offset_from_index<NumInvariantDim>(arg.outStrides_,
+                                                                                 invariant_index);
+
+                        if(!float_equal_zero{}(arg.beta_))
+                            accuVal += type_convert<AccDataType>(arg.out_host_[dst_offset]) *
+                                       type_convert<AccDataType>(arg.beta_);
+
+                        arg.out_host_[dst_offset]       = type_convert<OutDataType>(accuVal);
+                        arg.out_index_host_[dst_offset] = accuIndex;
+                    };
+
+                    std::size_t num_thread = std::thread::hardware_concurrency();
+
+                    std::size_t work_per_thread =
+                        (arg.invariant_index_set_.size() + num_thread - 1) / num_thread;
+
+                    std::vector<joinable_thread> threads(num_thread);
+
+                    for(std::size_t it = 0; it < num_thread; ++it)
+                    {
+                        std::size_t i_begin = it * work_per_thread;
+                        std::size_t i_end =
+                            std::min((it + 1) * work_per_thread, arg.invariant_index_set_.size());
+
+                        auto f = [=] {
+                            for(std::size_t i = i_begin; i < i_end; i++)
+                            {
+                                thread_reduce_func(arg.invariant_index_set_[i]);
+                            }
+                        };
+
+                        threads[it] = joinable_thread(f);
+                    }
+                };
+            }
+            else
+            {
+                using Accumulation =
+                    ck::detail::AccumulateWithNanCheck<PropagateNan, ReduceOperation, AccDataType>;
+
+                if constexpr(NumInvariantDim == 0)
+                {
+                    AccDataType accuVal = ReduceOperation::template GetIdentityValue<AccDataType>();
+
+                    for(const auto& reduce_index : arg.reduce_index_set_)
+                    {
+                        auto in_offset = get_offset_from_index<NumReduceDim>(arg.in_reduce_strides_,
+                                                                             reduce_index);
+
+                        auto currVal = type_convert<AccDataType>(arg.in_host_[in_offset]);
+
+                        arg.in_elementwise_op_(currVal, currVal);
+
+                        Accumulation::Calculate(accuVal, currVal);
+                    };
+
+                    arg.acc_elementwise_op_(accuVal, accuVal);
+
+                    if(!float_equal_one{}(arg.alpha_))
+                        accuVal *= type_convert<AccDataType>(arg.alpha_);
+
+                    if(!float_equal_zero{}(arg.beta_))
+                        accuVal += type_convert<AccDataType>(arg.out_host_[0]) *
+                                   type_convert<AccDataType>(arg.beta_);
+
+                    arg.out_host_[0] = type_convert<OutDataType>(accuVal);
+                }
+                else
+                {
+                    auto thread_reduce_func = [&](auto invariant_index) {
+                        AccDataType accuVal =
+                            ReduceOperation::template GetIdentityValue<AccDataType>();
+
+                        auto in_invariant_offset = get_offset_from_index<NumInvariantDim>(
+                            arg.in_invariant_strides_, invariant_index);
+
+                        for(const auto& reduce_index : arg.reduce_index_set_)
+                        {
+                            auto in_reduce_offset = get_offset_from_index<NumReduceDim>(
+                                arg.in_reduce_strides_, reduce_index);
+
+                            auto currVal = type_convert<AccDataType>(
+                                arg.in_host_[in_invariant_offset + in_reduce_offset]);
+
+                            arg.in_elementwise_op_(currVal, currVal);
+
+                            Accumulation::Calculate(accuVal, currVal);
+                        };
+
+                        arg.acc_elementwise_op_(accuVal, accuVal);
+
+                        if(!float_equal_one{}(arg.alpha_))
+                            accuVal *= type_convert<AccDataType>(arg.alpha_);
+
+                        auto dst_offset = get_offset_from_index<NumInvariantDim>(arg.outStrides_,
+                                                                                 invariant_index);
+
+                        if(!float_equal_zero{}(arg.beta_))
+                            accuVal += type_convert<AccDataType>(arg.out_host_[dst_offset]) *
+                                       type_convert<AccDataType>(arg.beta_);
+
+                        arg.out_host_[dst_offset] = type_convert<OutDataType>(accuVal);
+                    };
+
+                    std::size_t num_thread = std::thread::hardware_concurrency();
+
+                    std::size_t work_per_thread =
+                        (arg.invariant_index_set_.size() + num_thread - 1) / num_thread;
+
+                    std::vector<joinable_thread> threads(num_thread);
+
+                    for(std::size_t it = 0; it < num_thread; ++it)
+                    {
+                        std::size_t i_begin = it * work_per_thread;
+                        std::size_t i_end =
+                            std::min((it + 1) * work_per_thread, arg.invariant_index_set_.size());
+
+                        auto f = [=] {
+                            for(std::size_t i = i_begin; i < i_end; i++)
+                            {
+                                thread_reduce_func(arg.invariant_index_set_[i]);
+                            }
+                        };
+
+                        threads[it] = joinable_thread(f);
+                    }
+                };
+            };
+
+            return (0.0f);
+        };
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        };
+    };
+
+    bool IsSupportedArgument(const device::BaseArgument* p_arg) override
+    {
+        ignore = p_arg;
+
+        return true;
+    };
+
+    std::unique_ptr<device::BaseArgument>
+    MakeArgumentPointer(const std::array<index_t, Rank> inLengths,
+                        const std::array<index_t, Rank> inStrides,
+                        const std::array<index_t, NumDstDim> outLengths,
+                        const std::array<index_t, NumDstDim> outStrides,
+                        const std::array<int, NumReduceDim> reduceDims,
+                        double alpha,
+                        double beta,
+                        const void* in_host,
+                        const void* in_index_host,
+                        void* out_host,
+                        void* out_index_host,
+                        const InElementwiseOperation in_elementwise_op,
+                        const AccElementwiseOperation acc_elementwise_op) override
+    {
+        ignore = in_index_host;
+
+        return std::make_unique<Argument>(inLengths,
+                                          inStrides,
+                                          outLengths,
+                                          outStrides,
+                                          reduceDims,
+                                          alpha,
+                                          beta,
+                                          static_cast<const InDataType*>(in_host),
+                                          static_cast<OutDataType*>(out_host),
+                                          static_cast<IndexDataType*>(out_index_host),
+                                          in_elementwise_op,
+                                          acc_elementwise_op);
+    };
+
+    std::unique_ptr<device::BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>();
+    };
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "Reference_Reduce<" << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp
index 2882aa2203586a4852e12c39d065f23f57647b86..81071bcb0bccaf9cc6b2600337e810429f8632eb 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -24,17 +24,19 @@ struct ReferenceSoftmax : public device::BaseOperator
     {
         Argument(const Tensor<InDataType>& in,
                  Tensor<OutDataType>& out,
-                 AccDataType alpha,
-                 AccDataType beta,
+                 double alpha,
+                 double beta,
                  const std::vector<index_t> sm_reduce_dims,
                  Tensor<AccDataType>* sm_stats_ptr = nullptr)
             : in_(in),
               out_(out),
-              alpha_(alpha),
-              beta_(beta),
               sm_reduce_dims_(sm_reduce_dims),
               sm_stats_ptr_(sm_stats_ptr)
         {
+            alpha_ = static_cast<AccDataType>(alpha);
+            beta_  = static_cast<AccDataType>(beta);
+
+            // std::cout << "debug: scalar dims: ";  
             for(size_t i = 0; i < in.mDesc.GetNumOfDimension(); i++)
             {
                 if(std::find(sm_reduce_dims.begin(), sm_reduce_dims.end(), i) ==
@@ -171,8 +173,8 @@ struct ReferenceSoftmax : public device::BaseOperator
 
     static auto MakeArgument(const Tensor<InDataType>& in,
                              Tensor<OutDataType>& out,
-                             AccDataType alpha,
-                             AccDataType beta,
+                             double alpha,
+                             double beta,
                              const std::vector<index_t> sm_reduce_dims,
                              Tensor<AccDataType>* stats = nullptr)
     {
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp
index b6a9b0fb5ee4a3135750dad0d0818a0f6e439f3a..f949260ca4552b83ab6e8edafee9190d4f95cc64 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp b/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp
index df4fca6562755012200ff888677bcac1c90129a2..0b7887efb6e05f34b83a5bf684b5fedd3d78c385 100644
--- a/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef NAIVE_CONV_FWD_HPP
 #define NAIVE_CONV_FWD_HPP
diff --git a/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp b/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp
index 20df1b3616a016029f4083bd821b4e286b7b727c..f57fed9c07c7843af0f77db7991ce077b3021245 100644
--- a/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index 91980a9a66c6f57cdb8557f8ee6306abf31bf77d..851d9e497ffe6a0f7b506b43be841ccb8554bd6e 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -26,6 +26,7 @@ using Empty_Tuple = ck::Tuple<>;
 using F16_Tuple     = ck::Tuple<F16>;
 using F16_F16_Tuple = ck::Tuple<F16, F16>;
 
+using F64_Tuple     = ck::Tuple<F64>;
 using F32_Tuple     = ck::Tuple<F32>;
 using I32_Tuple     = ck::Tuple<I32>;
 using I32_F32_Tuple = ck::Tuple<I32, F32>;
@@ -85,11 +86,17 @@ using GK_GK_Tuple = ck::Tuple<GK, GK>;
 // pointwise functor
 using PassThrough    = ck::tensor_operation::element_wise::PassThrough;
 using Relu           = ck::tensor_operation::element_wise::Relu;
+using TanH           = ck::tensor_operation::element_wise::TanH;
 using Scale          = ck::tensor_operation::element_wise::Scale;
 using Bilinear       = ck::tensor_operation::element_wise::Bilinear;
 using AddAddFastGelu = ck::tensor_operation::element_wise::AddAddFastGelu;
 using AddFastGelu    = ck::tensor_operation::element_wise::AddFastGelu;
+using AddReluAdd     = ck::tensor_operation::element_wise::AddReluAdd;
 using FastGelu       = ck::tensor_operation::element_wise::FastGelu;
+using AddMultiply    = ck::tensor_operation::element_wise::AddMultiply;
+using ScaleAdd       = ck::tensor_operation::element_wise::ScaleAdd;
+using Gelu           = ck::tensor_operation::element_wise::Gelu;
+using Swish          = ck::tensor_operation::element_wise::Swish;
 
 template <typename Activation>
 using Activation_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<Activation>;
@@ -98,6 +105,10 @@ template <typename Activation>
 using Add_Activation_Mul_Clamp =
     ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<Activation>;
 
+template <typename Activation>
+using Add_Mul_Activation_Mul_Clamp =
+    ck::tensor_operation::element_wise::Add_Mul_Activation_Mul_Clamp<Activation>;
+
 template <typename Activation>
 using Activation_Mul2_Clamp = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<Activation>;
 
@@ -105,6 +116,10 @@ template <typename Activation>
 using Add_Activation_Mul2_Clamp =
     ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<Activation>;
 
+template <typename Activation>
+using Add_Mul2_Activation_Mul_Clamp =
+    ck::tensor_operation::element_wise::Add_Mul2_Activation_Mul_Clamp<Activation>;
+
 template <typename DeviceOp, typename Tag = void>
 struct DeviceOperationInstanceFactory;
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
index 0655fd92e44c5c6400c5a66e6ee9e21f6e50da71..c3c8c0e5a76329071970a2ee9e81f297fbf11292 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
index 495c5f884fd633898867a3e0a5fc121326527c8e..73f6004257135b75ac0bb1cd79f92e0d9d78c3ef 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..70bff278997501254d47595f28cb0d2d01c679e2
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_batched_contraction_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_mnnm_instance(
+    std::vector<std::unique_ptr<
+        DeviceBatchedContractionMultipleD<1,
+                                          2,
+                                          3,
+                                          1,
+                                          F16,
+                                          F16,
+                                          F16_Tuple,
+                                          F16,
+                                          ck::tensor_operation::element_wise::PassThrough,
+                                          ck::tensor_operation::element_wise::PassThrough,
+                                          ck::tensor_operation::element_wise::Add>>>& instances);
+
+// Contraction + add
+template <index_t NumDimG,
+          index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename DDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceBatchedContractionMultipleD<
+        NumDimG,
+        NumDimM,
+        NumDimN,
+        NumDimK,
+        ADataType,
+        BDataType,
+        ck::Tuple<DDataType>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::Add>>
+{
+    using DeviceOp =
+        DeviceBatchedContractionMultipleD<NumDimG,
+                                          NumDimM,
+                                          NumDimN,
+                                          NumDimK,
+                                          ADataType,
+                                          BDataType,
+                                          ck::Tuple<DDataType>,
+                                          EDataType,
+                                          ck::tensor_operation::element_wise::PassThrough,
+                                          ck::tensor_operation::element_wise::PassThrough,
+                                          ck::tensor_operation::element_wise::Add>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::half_t> &&
+                     is_same_v<DDataType, ck::half_t> && is_same_v<EDataType, ck::half_t>)
+        {
+            if constexpr(NumDimG == 1 && NumDimM == 2 && NumDimN == 3 && NumDimK == 1)
+            {
+                add_device_batched_contraction_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_mnnm_instance(
+                    op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..33653a30800ebb7a22d333d0ccc1bd8ed4506df4
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
@@ -0,0 +1,190 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
+    std::vector<std::unique_ptr<
+        DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                            1,
+                                            1,
+                                            1,
+                                            1,
+                                            F16,
+                                            F16,
+                                            F16,
+                                            F16,
+                                            ck::Tuple<F16>,
+                                            ck::Tuple<>,
+                                            PassThrough,
+                                            PassThrough,
+                                            ScaleAdd,
+                                            PassThrough,
+                                            PassThrough,
+                                            MaskingSpecialization::MaskOutUpperTriangle>>>&
+        instances);
+
+void add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
+    std::vector<
+        std::unique_ptr<DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            F16,
+                                                            F16,
+                                                            F16,
+                                                            F16,
+                                                            ck::Tuple<F16>,
+                                                            ck::Tuple<>,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            ScaleAdd,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            MaskingSpecialization::MaskDisabled>>>&
+        instances);
+
+void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
+    std::vector<std::unique_ptr<
+        DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                            1,
+                                            1,
+                                            1,
+                                            1,
+                                            BF16,
+                                            BF16,
+                                            BF16,
+                                            BF16,
+                                            ck::Tuple<BF16>,
+                                            ck::Tuple<>,
+                                            PassThrough,
+                                            PassThrough,
+                                            ScaleAdd,
+                                            PassThrough,
+                                            PassThrough,
+                                            MaskingSpecialization::MaskOutUpperTriangle>>>&
+        instances);
+
+void add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
+    std::vector<
+        std::unique_ptr<DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            BF16,
+                                                            BF16,
+                                                            BF16,
+                                                            BF16,
+                                                            ck::Tuple<BF16>,
+                                                            ck::Tuple<>,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            ScaleAdd,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            MaskingSpecialization::MaskDisabled>>>&
+        instances);
+
+template <typename ADataType,
+          typename B0DataType,
+          typename B1DataType,
+          typename CDataType,
+          typename Acc0BiasDataType,
+          MaskingSpecialization MaskingSpec>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                                      1,
+                                                                      1,
+                                                                      1,
+                                                                      1,
+                                                                      ADataType,
+                                                                      B0DataType,
+                                                                      B1DataType,
+                                                                      CDataType,
+                                                                      Acc0BiasDataType,
+                                                                      ck::Tuple<>,
+                                                                      PassThrough,
+                                                                      PassThrough,
+                                                                      ScaleAdd,
+                                                                      PassThrough,
+                                                                      PassThrough,
+                                                                      MaskingSpec>>
+{
+    using DeviceOp = DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                         1,
+                                                         1,
+                                                         1,
+                                                         1,
+                                                         ADataType,
+                                                         B0DataType,
+                                                         B1DataType,
+                                                         CDataType,
+                                                         Acc0BiasDataType,
+                                                         ck::Tuple<>,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         ScaleAdd,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         MaskingSpec>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<B0DataType, half_t> &&
+                     is_same_v<B1DataType, half_t> && is_same_v<CDataType, half_t> &&
+                     Acc0BiasDataType::Size() == 1 &&
+                     is_same_v<tuple_element_t<0, Acc0BiasDataType>, half_t>)
+        {
+            if constexpr(MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle)
+            {
+                add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
+                    op_ptrs);
+            }
+            else if(MaskingSpec == MaskingSpecialization::MaskDisabled)
+            {
+                add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
+                    op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<ADataType, BF16> && is_same_v<B0DataType, BF16> &&
+                          is_same_v<B1DataType, BF16> && is_same_v<CDataType, BF16> &&
+                          Acc0BiasDataType::Size() == 1 &&
+                          is_same_v<tuple_element_t<0, Acc0BiasDataType>, BF16>)
+        {
+            if constexpr(MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle)
+            {
+                add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
+                    op_ptrs);
+            }
+            else if(MaskingSpec == MaskingSpecialization::MaskDisabled)
+            {
+                add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
+                    op_ptrs);
+            }
+        }
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
index a6dcfa30d3efd2d78562c1829ffb72b4eaabca9b..28ccf61a3a1d46f3dfb6c7158192416622f7907a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..3d50aae09238523875c6f8ff8d1abd8f18678185
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
@@ -0,0 +1,337 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatchedGemmMultiD<
+    ALayout,
+    BLayout,
+    Empty_Tuple,
+    ELayout,
+    ADataType,
+    BDataType,
+    Empty_Tuple,
+    EDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough>>
+{
+    using DeviceOp = DeviceBatchedGemmMultiD<ALayout,
+                                             BLayout,
+                                             Empty_Tuple,
+                                             ELayout,
+                                             ADataType,
+                                             BDataType,
+                                             Empty_Tuple,
+                                             EDataType,
+                                             ck::tensor_operation::element_wise::PassThrough,
+                                             ck::tensor_operation::element_wise::PassThrough,
+                                             ck::tensor_operation::element_wise::PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
+                          is_same_v<EDataType, int8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
index 8a0b1b1fa7de54954a5bc375241bb7a0686b2e49..0a30c210dd818b7812a0ae52690afb5a7a991ea2 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
index 89df1a7a0d44394d190973ad787b176da8002851..2ff64675c904f0f6e30c436e6c44b57211c00f89 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
index c84ffcff8cbb8d4820f88dbfa3e7de6a04961bb8..0e1f6f04e8c58812651d3d549601952050fb73dd 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
index 8e40d60c17b8dc8d743e29cc1472a1e9eed9bdc1..8fd1c7665d143e9a0fb07f47f07ead18aa33aa88 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..f6f4df7e2ec0e34b6ac7a4b356649289f6a30c91
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
@@ -0,0 +1,117 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/device_elementwise.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// FP16
+void add_device_batchnorm_infer_rank_4_f16_instances(
+    std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<F16, F32, F32, F16, F16>,
+        ck::Tuple<F16>,
+        ck::tensor_operation::element_wise::NormalizeInInfer,
+        4>>>&);
+
+// FP32
+void add_device_batchnorm_infer_rank_4_f32_instances(
+    std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<F32, F32, F32, F32, F32>,
+        ck::Tuple<F32>,
+        ck::tensor_operation::element_wise::NormalizeInInfer,
+        4>>>&);
+
+// BF16
+void add_device_batchnorm_infer_rank_4_bf16_instances(
+    std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<BF16, F32, F32, BF16, BF16>,
+        ck::Tuple<BF16>,
+        ck::tensor_operation::element_wise::NormalizeInInfer,
+        4>>>&);
+
+// FP64
+void add_device_batchnorm_infer_rank_4_f64_instances(
+    std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<F64, F64, F64, F64, F64>,
+        ck::Tuple<F64>,
+        ck::tensor_operation::element_wise::NormalizeInInfer,
+        4>>>&);
+
+template <typename XDataType,
+          typename YDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          index_t Rank>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElementwise<
+    ck::Tuple<XDataType, MeanVarDataType, MeanVarDataType, ScaleDataType, BiasDataType>,
+    ck::Tuple<YDataType>,
+    ck::tensor_operation::element_wise::NormalizeInInfer,
+    Rank>>
+{
+    using DeviceOp = ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<XDataType, MeanVarDataType, MeanVarDataType, ScaleDataType, BiasDataType>,
+        ck::Tuple<YDataType>,
+        ck::tensor_operation::element_wise::NormalizeInInfer,
+        Rank>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<XDataType, F16> && is_same_v<YDataType, F16> &&
+                     is_same_v<ScaleDataType, F16> && is_same_v<BiasDataType, F16> &&
+                     is_same_v<MeanVarDataType, F32>)
+        {
+            if constexpr(Rank == 4)
+            {
+                add_device_batchnorm_infer_rank_4_f16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
+                          is_same_v<ScaleDataType, F32> && is_same_v<BiasDataType, F32> &&
+                          is_same_v<MeanVarDataType, F32>)
+        {
+            if constexpr(Rank == 4)
+            {
+                add_device_batchnorm_infer_rank_4_f32_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
+                          is_same_v<ScaleDataType, BF16> && is_same_v<BiasDataType, BF16> &&
+                          is_same_v<MeanVarDataType, F32>)
+        {
+            if constexpr(Rank == 4)
+            {
+                add_device_batchnorm_infer_rank_4_bf16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
+                          is_same_v<ScaleDataType, F64> && is_same_v<BiasDataType, F64> &&
+                          is_same_v<MeanVarDataType, F64>)
+        {
+            if constexpr(Rank == 4)
+            {
+                add_device_batchnorm_infer_rank_4_f64_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
index a0cea7e390aab9402db40a439ffb43ccfb335bde..2ed8255f6d66b72cba17492351824a0007afe6dc 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
@@ -1,12 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
@@ -19,6 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+// float
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -67,6 +66,55 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            PassThrough,
                                                            Bilinear>>>& instances);
 
+// double
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances);
+
 // Contraction + Bilinear
 template <index_t NumDimM,
           index_t NumDimN,
@@ -118,6 +166,22 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
             }
         }
 
+        if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
+                     is_same_v<DDataType, double> && is_same_v<EDataType, double>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
+                    op_ptrs);
+            }
+        }
+
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
index e921ecd47aa5b08d2f6f63ee2bad30c40608b05c..5d9567731dae2ea2d491cbeade861486f36a1743 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
@@ -1,12 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
@@ -19,6 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+// float
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -67,6 +66,55 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            PassThrough,
                                                            Scale>>>& instances);
 
+// double
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances);
+
 // Contraction + Scale
 template <index_t NumDimM,
           index_t NumDimN,
@@ -117,6 +165,22 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
             }
         }
 
+        if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
+                     is_same_v<EDataType, double>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
+                    op_ptrs);
+            }
+        }
+
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
index ec5d18fc214c34c1e972aa5fc50726cd41f8180c..1efe073669846e40274897ec33cfbf932cb31715 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
index 62f28c9b11d287f355d5feeeb24bf7863ac96851..dd8c1987cfe65bfba4e792f269004f52cdd912ad 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
index 141af558471852471833459edb1b9a07e1108069..b03693b00aabceef8614a56e91c0ce9b89207def 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
@@ -1,16 +1,14 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/device_elementwise.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -18,11 +16,8 @@ namespace device {
 namespace instance {
 
 using Normalize                            = ck::tensor_operation::element_wise::Normalize;
-using DeviceNormalizeFromMeanMeanSquarePtr = ck::tensor_operation::device::DeviceElementwiseBasePtr<
-    Tuple<half_t, float, float, half_t, half_t>,
-    Tuple<half_t>,
-    Normalize,
-    2>;
+using DeviceNormalizeFromMeanMeanSquarePtr = ck::tensor_operation::device::
+    DeviceElementwisePtr<Tuple<half_t, float, float, half_t, half_t>, Tuple<half_t>, Normalize, 2>;
 
 void add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(
     std::vector<DeviceNormalizeFromMeanMeanSquarePtr>& instances);
@@ -33,20 +28,34 @@ template <typename InputType,
           typename GammaDataType,
           typename BetaDataType,
           typename OutputType>
-auto get_device_normalize_from_mean_meansquare_instances()
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElementwise<
+    ck::Tuple<InputType, MeanType, MeanSquareType, GammaDataType, BetaDataType>,
+    ck::Tuple<OutputType>,
+    Normalize,
+    2>>
 {
-    std::vector<DeviceNormalizeFromMeanMeanSquarePtr> op_ptrs;
+    using DeviceOp = DeviceElementwise<
+        ck::Tuple<InputType, MeanType, MeanSquareType, GammaDataType, BetaDataType>,
+        ck::Tuple<OutputType>,
+        Normalize,
+        2>;
 
-    if constexpr(is_same<InputType, half_t>::value && is_same<MeanType, float>::value &&
-                 is_same<MeanSquareType, float>::value && is_same<GammaDataType, half_t>::value &&
-                 is_same<BetaDataType, half_t>::value && is_same<OutputType, half_t>::value)
+    static auto GetInstances()
     {
-        ck::tensor_operation::device::instance::
-            add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(op_ptrs);
-    }
-
-    return op_ptrs;
-}
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same<InputType, half_t>::value && is_same<MeanType, float>::value &&
+                     is_same<MeanSquareType, float>::value &&
+                     is_same<GammaDataType, half_t>::value &&
+                     is_same<BetaDataType, half_t>::value && is_same<OutputType, half_t>::value)
+        {
+            ck::tensor_operation::device::instance::
+                add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(op_ptrs);
+        }
+
+        return op_ptrs;
+    };
+};
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
index 682f54675982e9338823a814c46075242534de1d..b15139510935586a930cd5a7c5a091299dfddecc 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include <cstdlib>
-
+#include <vector>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_reduce.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
index c87ae159bee2635c903490b9e09096be8c242c8c..a1c006cf62a063a742a8dd29365a0aefe359282d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index e230507e7e30a605bac4ed5d27cea454739ba24a..adac7d0dcf4a7bee2a0315add52d4c85251208dc 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -1,12 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
 #include <memory>
 #include <vector>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm.hpp"
@@ -24,21 +22,41 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -65,21 +83,41 @@ void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -297,6 +335,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
@@ -304,6 +343,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
             }
@@ -312,6 +352,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
@@ -319,6 +360,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
@@ -354,24 +396,28 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(op_ptrs);
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
index 09d8e8b95bb28f393e8d9b467363b0ccb78a1124..99b2ad13152f0baa82872add5efdeb9b6a1988a8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
@@ -1,16 +1,13 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
index 554437f4903bcae96f68f2a53097a6eaf88367f3..fd3550c2f01b733602e6ae84dcca34f74255c9fe 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..481915d00b7416a444eabdf92289d275a9f60ef9
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>&);
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>&);
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>&);
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>&);
+
+// GEMM + Add + Multiply
+template <typename ALayout,
+          typename BLayout,
+          typename D0Layout,
+          typename D1Layout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename D0DataType,
+          typename D1DataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMultipleD<
+    ALayout,
+    BLayout,
+    ck::Tuple<D0Layout, D1Layout>,
+    ELayout,
+    ADataType,
+    BDataType,
+    ck::Tuple<D0DataType, D1DataType>,
+    EDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::AddMultiply>>
+{
+    using DeviceOp = DeviceGemmMultipleD<ALayout,
+                                         BLayout,
+                                         ck::Tuple<D0Layout, D1Layout>,
+                                         ELayout,
+                                         ADataType,
+                                         BDataType,
+                                         ck::Tuple<D0DataType, D1DataType>,
+                                         EDataType,
+                                         ck::tensor_operation::element_wise::PassThrough,
+                                         ck::tensor_operation::element_wise::PassThrough,
+                                         ck::tensor_operation::element_wise::AddMultiply>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<D0DataType, half_t> && is_same_v<D1DataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..de21f325a6711d1fe746905dfb7686ddb4ff7e7c
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
@@ -0,0 +1,172 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Row,
+                                                             Row,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>&);
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Row,
+                                                             Col,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>&);
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Col,
+                                                             Row,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>&);
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Col,
+                                                             Col,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>&);
+
+// GEMM + Add + Relu + Add + Layernorm
+template <typename ALayout,
+          typename BLayout,
+          typename D0Layout,
+          typename D1Layout,
+          typename HLayout,
+          typename ADataType,
+          typename BDataType,
+          typename D0DataType,
+          typename D1DataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename HDataType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMultipleDLayernorm<
+    ALayout,
+    BLayout,
+    ck::Tuple<D0Layout, D1Layout>,
+    HLayout,
+    ADataType,
+    BDataType,
+    ck::Tuple<D0DataType, D1DataType>,
+    GammaDataType,
+    BetaDataType,
+    HDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::AddReluAdd,
+    ck::tensor_operation::element_wise::PassThrough>>
+{
+    using DeviceOp = DeviceGemmMultipleDLayernorm<ALayout,
+                                                  BLayout,
+                                                  ck::Tuple<D0Layout, D1Layout>,
+                                                  HLayout,
+                                                  ADataType,
+                                                  BDataType,
+                                                  ck::Tuple<D0DataType, D1DataType>,
+                                                  GammaDataType,
+                                                  BetaDataType,
+                                                  HDataType,
+                                                  ck::tensor_operation::element_wise::PassThrough,
+                                                  ck::tensor_operation::element_wise::PassThrough,
+                                                  ck::tensor_operation::element_wise::AddReluAdd,
+                                                  ck::tensor_operation::element_wise::PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<D0DataType, half_t> && is_same_v<D1DataType, half_t> &&
+                     is_same_v<GammaDataType, half_t> && is_same_v<BetaDataType, half_t> &&
+                     is_same_v<HDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                         is_same_v<HLayout, Row>)
+            {
+                add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<HLayout, Row>)
+            {
+                add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<HLayout, Row>)
+            {
+                add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<HLayout, Row>)
+            {
+                add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
index ef70504f29b717ed70a494926d256f8b98a35e47..f80efd5511a379a300078bd8d2cb63bcc0ce0154 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
@@ -1,12 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
index fbc5df98a4ec116976d491835e8a8411b609c992..09b1c2190facdd84e09e5e1d66f32b4212ef0ec8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index 8986a793444c8f73679d892f4d55adb02bc61834..534875151105ee04a147584e61623320ee18ce57 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_splitk.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0f44952810c667cb101bdc0686d30a3d983e5785
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
@@ -0,0 +1,119 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::tensor_layout::convolution;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdWeightDefault =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
+
+static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances = std::tuple<
+    // clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances = std::tuple<
+    // clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances = std::tuple<
+    // clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
index 81b2b4fcf37cb1f0d05415eb7d56587f0a29f380..0b20e19a2a980975d057798549055d2c4a8d0ef0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -30,6 +30,76 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
@@ -78,6 +148,35 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
             }
+            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                              is_same_v<OutDataType, F32>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                              is_same_v<OutDataType, BF16>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
+        {
+            if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                         is_same_v<OutDataType, F16>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                              is_same_v<OutDataType, F32>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                              is_same_v<OutDataType, BF16>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
+            }
         }
 
         return op_ptrs;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index ef6920e52a2096a3f61e40435c6fa3ce64a17323..d0c3338f245932de82d99f0fc8188ef36b555a10 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -91,6 +91,42 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
 // conv3d backward weight
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
@@ -162,66 +198,103 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC> &&
-                     is_same_v<WeiLayout, GKXC> && is_same_v<OutLayout, GNWK>)
+        if constexpr(NumDimSpatial == 1)
         {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
-            {
-                add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
-            {
-                add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+            if constexpr(is_same_v<InLayout, GNWC> && is_same_v<WeiLayout, GKXC> &&
+                         is_same_v<OutLayout, GNWK>)
             {
-                add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
-                    op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
-                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
+        else if constexpr(NumDimSpatial == 2)
         {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
-            {
-                add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+            if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
+                         is_same_v<OutLayout, GNHWK>)
             {
-                add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+            else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
+                              is_same_v<OutLayout, NHWGK>)
             {
-                add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
-                    op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
-                          is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, GNDHWK>)
+        else if constexpr(NumDimSpatial == 3)
         {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
-            {
-                add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
-                    op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
-            {
-                add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
-                    op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+            if(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
+               is_same_v<OutLayout, GNDHWK>)
             {
-                add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
-                    op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index ee38b738274c5492f27ed0e7ad5c285eacf1f54b..627a5ae2aa9d49d1e9936d094f4240eb9907c304 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include <vector>
-
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
@@ -117,20 +117,6 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -159,20 +145,21 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances(
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
+                                                              NHWGK,
+                                                              BF16,
+                                                              BF16,
                                                               Empty_Tuple,
-                                                              int8_t,
+                                                              BF16,
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -187,6 +174,20 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
 // grouped conv3d forward, GNDHWC/GKZYXC/GNDHWK
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -244,6 +245,63 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+// grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              BF16,
+                                                              BF16,
+                                                              Empty_Tuple,
+                                                              BF16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -328,12 +386,6 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
-            {
-                add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances(op_ptrs);
-                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances(op_ptrs);
-            }
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                           is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
@@ -341,7 +393,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
-                // no instance
+                add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                               is_same_v<OutDataType, half_t>)
@@ -352,12 +404,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                               is_same_v<WeiDataType, ck::bhalf_t> &&
                               is_same_v<OutDataType, ck::bhalf_t>)
             {
-                // no instance
-            }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
-            {
-                // no instance
+                add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
             }
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
@@ -385,6 +432,31 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(op_ptrs);
             }
         }
+        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
+                          is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
+        {
+            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                         is_same_v<OutDataType, float>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                              is_same_v<OutDataType, half_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                              is_same_v<WeiDataType, ck::bhalf_t> &&
+                              is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                              is_same_v<OutDataType, int8_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(op_ptrs);
+            }
+        }
 
         return op_ptrs;
     }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
index c64598daddb6874bd9c01eba6045b42c91792d14..b482e97ee0386ba8d7940481f93512e797aa2f4f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
@@ -68,24 +68,75 @@ void add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
 template <typename ALayout,
           typename BLayout,
           typename ELayout,
           typename ADataType,
           typename BDataType,
           typename EDataType>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedGemm<
-    ALayout,
-    BLayout,
-    Empty_Tuple,
-    ELayout,
-    ADataType,
-    BDataType,
-    Empty_Tuple,
-    EDataType,
-    ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough>>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedGemm<ALayout,
+                                                                                      BLayout,
+                                                                                      Empty_Tuple,
+                                                                                      ELayout,
+                                                                                      ADataType,
+                                                                                      BDataType,
+                                                                                      Empty_Tuple,
+                                                                                      EDataType,
+                                                                                      PassThrough,
+                                                                                      PassThrough,
+                                                                                      PassThrough>>
 {
     using DeviceOp = DeviceGroupedGemm<ALayout,
                                        BLayout,
@@ -95,9 +146,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                        BDataType,
                                        Empty_Tuple,
                                        EDataType,
-                                       ck::tensor_operation::element_wise::PassThrough,
-                                       ck::tensor_operation::element_wise::PassThrough,
-                                       ck::tensor_operation::element_wise::PassThrough>;
+                                       PassThrough,
+                                       PassThrough,
+                                       PassThrough>;
 
     static auto GetInstances()
     {
@@ -110,11 +161,17 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<ELayout, Row>)
             {
                 add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
+                    op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<ELayout, Row>)
             {
                 add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
+                    op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<ELayout, Row>)
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fastgelu.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..eb181ebc7f24dfbc6ec9886767d7c2aaea57f240
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fastgelu.hpp
@@ -0,0 +1,136 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <memory>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances);
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances);
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances);
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances);
+
+// GroupedGEMM + GELU
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedGemm<ALayout,
+                                                                                      BLayout,
+                                                                                      Empty_Tuple,
+                                                                                      ELayout,
+                                                                                      ADataType,
+                                                                                      BDataType,
+                                                                                      Empty_Tuple,
+                                                                                      EDataType,
+                                                                                      PassThrough,
+                                                                                      PassThrough,
+                                                                                      FastGelu>>
+{
+    using DeviceOp = DeviceGroupedGemm<ALayout,
+                                       BLayout,
+                                       Empty_Tuple,
+                                       ELayout,
+                                       ADataType,
+                                       BDataType,
+                                       Empty_Tuple,
+                                       EDataType,
+                                       PassThrough,
+                                       PassThrough,
+                                       FastGelu>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+            }
+        }
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
index 55c67b7623b85c03381833431338f6be083fbd7b..778f625d84cbba947bd4d81ecc59bb94ebfe195b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_normalization.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..23917752997b42e7c67e667247e38a1762a50a8d
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_normalization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// FP16
+void add_device_normalization_rank_5_3_swish_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Swish, 5, 3>>>&);
+
+// FP32
+void add_device_normalization_rank_5_3_swish_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Swish, 5, 3>>>&);
+
+// [x, gamma, beta, y] = [f16, f32, f32, f16]
+void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F32, F16, Swish, 5, 3>>>&);
+
+template <typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          index_t Rank,
+          index_t NumReduceDim>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceNormalization<XDataType,
+                                                      GammaDataType,
+                                                      BetaDataType,
+                                                      F32,
+                                                      YDataType,
+                                                      ck::tensor_operation::element_wise::Swish,
+                                                      Rank,
+                                                      NumReduceDim>>
+{
+    using DeviceOp = DeviceNormalization<XDataType,
+                                         GammaDataType,
+                                         BetaDataType,
+                                         F32,
+                                         YDataType,
+                                         ck::tensor_operation::element_wise::Swish,
+                                         Rank,
+                                         NumReduceDim>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
+                     is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16>)
+        {
+            if constexpr(Rank == 5 && NumReduceDim == 3)
+            {
+                add_device_normalization_rank_5_3_swish_f16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
+                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
+        {
+            if constexpr(Rank == 5 && NumReduceDim == 3)
+            {
+                add_device_normalization_rank_5_3_swish_f32_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F32> &&
+                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F16>)
+        {
+            if constexpr(Rank == 5 && NumReduceDim == 3)
+            {
+                add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..ccb5cb5a9eb9e74932789cfa25f299707746bce6
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
@@ -0,0 +1,111 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto InOutRank  = 4;
+static constexpr auto WindowRank = 2;
+
+static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
+static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
+
+// FP16
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
+
+// FP16 - return index
+void add_device_pool2d_fwd_nhwc_index_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
+
+// FP32
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
+
+// FP32 - return index
+void add_device_pool2d_fwd_nhwc_index_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  OutputIndex>>
+{
+    using DeviceOp = DevicePoolFwd<InOutRank,
+                                   WindowRank,
+                                   InDataType,
+                                   OutDataType,
+                                   IndexDataType,
+                                   ReduceOpId,
+                                   OutputIndex>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
+                     is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool2d_fwd_nhwc_index_f16_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool2d_fwd_nhwc_f16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                          is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool2d_fwd_nhwc_index_f32_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool2d_fwd_nhwc_f32_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..3a006b00a5b04c2689d2a262aecee5623f02d195
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
@@ -0,0 +1,111 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto InOutRank  = 5;
+static constexpr auto WindowRank = 3;
+
+static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
+static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
+
+// FP16
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
+
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
+
+// FP16 - return index
+void add_device_pool3d_fwd_ndhwc_index_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
+
+// FP32
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
+
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
+
+// FP32 - return index
+void add_device_pool3d_fwd_ndhwc_index_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  OutputIndex>>
+{
+    using DeviceOp = DevicePoolFwd<InOutRank,
+                                   WindowRank,
+                                   InDataType,
+                                   OutDataType,
+                                   IndexDataType,
+                                   ReduceOpId,
+                                   OutputIndex>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
+                     is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool3d_fwd_ndhwc_index_f16_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool3d_fwd_ndhwc_f16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                          is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool3d_fwd_ndhwc_index_f32_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool3d_fwd_ndhwc_f32_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..2ed4b0d5f880d5e8f0e8ba26e7e2ba541201c63f
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
@@ -0,0 +1,224 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Col, Row, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+// Layout(A, B, C) = [Col, Col, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+// Layout(A, B, C) = [Row, Row, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+// Layout(A, B, C) = [Row, Col, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+// Layout(A, B, C) = [Col, Row, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+// Layout(A, B, C) = [Col, Col, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+// Layout(A, B, C) = [Row, Row, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+// Layout(A, B, C) = [Row, Col, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType,
+          typename Activation>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMultipleD<
+    ALayout,
+    BLayout,
+    Empty_Tuple,
+    ELayout,
+    ADataType,
+    BDataType,
+    Empty_Tuple,
+    EDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    Activation_Mul_Clamp<Activation>>>
+{
+    using DeviceOp = DeviceGemmMultipleD<ALayout,
+                                         BLayout,
+                                         Empty_Tuple,
+                                         ELayout,
+                                         ADataType,
+                                         BDataType,
+                                         Empty_Tuple,
+                                         EDataType,
+                                         ck::tensor_operation::element_wise::PassThrough,
+                                         ck::tensor_operation::element_wise::PassThrough,
+                                         Activation_Mul_Clamp<Activation>>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
+                     is_same_v<EDataType, int8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                if constexpr(is_same_v<Activation, PassThrough>)
+                {
+                    add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+                    add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                if constexpr(is_same_v<Activation, PassThrough>)
+                {
+                    add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+                    add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                if constexpr(is_same_v<Activation, PassThrough>)
+                {
+                    add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+                    add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                if constexpr(is_same_v<Activation, PassThrough>)
+                {
+                    add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+                    add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+                }
+            }
+
+            return op_ptrs;
+        }
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
index eda81a233c01c7c2bff96a873a3db823f49f572c..8a96f6707e35344762003a958bc7afb9b937c86f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,14 +17,14 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
-void add_device_conv2d_bias_perchannel_quantization_int8_instances(
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_F32_Tuple,
@@ -34,12 +34,12 @@ void add_device_conv2d_bias_perchannel_quantization_int8_instances(
                                                       Add_Activation_Mul2_Clamp<PassThrough>>>>&
         instances);
 
-void add_device_conv2d_bias_relu_perchannel_quantization_int8_instances(
+void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -49,6 +49,70 @@ void add_device_conv2d_bias_relu_perchannel_quantization_int8_instances(
                                                               Add_Activation_Mul2_Clamp<Relu>>>>&
         instances);
 
+void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                      NHWGC,
+                                                      GKYXC,
+                                                      GK_GK_Tuple,
+                                                      NHWGK,
+                                                      int8_t,
+                                                      int8_t,
+                                                      I32_F32_Tuple,
+                                                      int8_t,
+                                                      PassThrough,
+                                                      PassThrough,
+                                                      Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
+        instances);
+
+void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                      NHWGC,
+                                                      GKYXC,
+                                                      GK_GK_Tuple,
+                                                      NHWGK,
+                                                      int8_t,
+                                                      int8_t,
+                                                      I32_F32_Tuple,
+                                                      int8_t,
+                                                      PassThrough,
+                                                      PassThrough,
+                                                      Add_Activation_Mul2_Clamp<PassThrough>>>>&
+        instances);
+
+void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Activation_Mul2_Clamp<Relu>>>>&
+        instances);
+
+void add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                      NHWGC,
+                                                      GKYXC,
+                                                      GK_GK_Tuple,
+                                                      NHWGK,
+                                                      int8_t,
+                                                      int8_t,
+                                                      I32_F32_Tuple,
+                                                      int8_t,
+                                                      PassThrough,
+                                                      PassThrough,
+                                                      Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
+        instances);
+
+// piecewise activation function
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -90,17 +154,84 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<DsDataType, I32_F32_Tuple> && is_same_v<OutDataType, int8_t>)
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
-                    add_device_conv2d_bias_perchannel_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(op_ptrs);
+                }
                 else if constexpr(is_same_v<Activation, Relu>)
-                    add_device_conv2d_bias_relu_perchannel_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(op_ptrs);
+                }
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+// non-piecewise activation function
+template <ck::index_t NumDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename DsDataType,
+          typename OutDataType,
+          typename Activation>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+    NumDimSpatial,
+    InLayout,
+    WeiLayout,
+    DsLayout,
+    OutLayout,
+    InDataType,
+    WeiDataType,
+    DsDataType,
+    OutDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    Add_Mul2_Activation_Mul_Clamp<Activation>>>
+{
+    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
+                                                   InLayout,
+                                                   WeiLayout,
+                                                   DsLayout,
+                                                   OutLayout,
+                                                   InDataType,
+                                                   WeiDataType,
+                                                   DsDataType,
+                                                   OutDataType,
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Add_Mul2_Activation_Mul_Clamp<Activation>>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+                     is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_GK_Tuple> &&
+                     is_same_v<OutLayout, NHWGK>)
+        {
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<DsDataType, I32_F32_Tuple> && is_same_v<OutDataType, int8_t>)
+            {
+                if constexpr(is_same_v<Activation, TanH>)
+                {
+                    add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(op_ptrs);
+                }
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
index 113840263893688f2adfc17060a77428628ad6bd..e17bea5fdf05ba19bb9e36de450865b67c406ff7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,14 +17,14 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
-void add_device_conv2d_bias_perlayer_quantization_int8_instances(
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_Tuple,
@@ -34,12 +34,12 @@ void add_device_conv2d_bias_perlayer_quantization_int8_instances(
                                                       Add_Activation_Mul_Clamp<PassThrough>>>>&
         instances);
 
-void add_device_conv2d_bias_relu_perlayer_quantization_int8_instances(
+void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -49,6 +49,68 @@ void add_device_conv2d_bias_relu_perlayer_quantization_int8_instances(
                                                               Add_Activation_Mul_Clamp<Relu>>>>&
         instances);
 
+void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul_Activation_Mul_Clamp<TanH>>>>&
+        instances);
+
+void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
+    std::vector<
+        std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                      NHWGC,
+                                                      GKYXC,
+                                                      GK_Tuple,
+                                                      NHWGK,
+                                                      int8_t,
+                                                      int8_t,
+                                                      I32_Tuple,
+                                                      int8_t,
+                                                      PassThrough,
+                                                      PassThrough,
+                                                      Add_Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Activation_Mul_Clamp<Relu>>>>&
+        instances);
+
+void add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul_Activation_Mul_Clamp<TanH>>>>&
+        instances);
+
+// piecewise activation function
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -90,17 +152,84 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<DsDataType, I32_Tuple> && is_same_v<OutDataType, int8_t>)
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
-                    add_device_conv2d_bias_perlayer_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(op_ptrs);
+                }
                 else if constexpr(is_same_v<Activation, Relu>)
-                    add_device_conv2d_bias_relu_perlayer_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(op_ptrs);
+                }
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+// non-piecewise activation function
+template <ck::index_t NumDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename DsDataType,
+          typename OutDataType,
+          typename Activation>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
+    NumDimSpatial,
+    InLayout,
+    WeiLayout,
+    DsLayout,
+    OutLayout,
+    InDataType,
+    WeiDataType,
+    DsDataType,
+    OutDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    Add_Mul_Activation_Mul_Clamp<Activation>>>
+{
+    using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
+                                                   InLayout,
+                                                   WeiLayout,
+                                                   DsLayout,
+                                                   OutLayout,
+                                                   InDataType,
+                                                   WeiDataType,
+                                                   DsDataType,
+                                                   OutDataType,
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Add_Mul_Activation_Mul_Clamp<Activation>>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+                     is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_Tuple> &&
+                     is_same_v<OutLayout, NHWGK>)
+        {
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<DsDataType, I32_Tuple> && is_same_v<OutDataType, int8_t>)
+            {
+                if constexpr(is_same_v<Activation, TanH>)
+                {
+                    add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(op_ptrs);
+                }
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
index 1a67ce56880d5ce21f0116ce25a08b53c474fc6d..9236b5c79e3079b0ee1f7c68091f15866d692a3a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,13 +17,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
-void add_device_conv2d_perchannel_quantization_int8_instances(
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+void add_device_conv2d_dl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -33,12 +33,42 @@ void add_device_conv2d_perchannel_quantization_int8_instances(
                                                               Activation_Mul2_Clamp<PassThrough>>>>&
         instances);
 
-void add_device_conv2d_relu_perchannel_quantization_int8_instances(
+void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Activation_Mul2_Clamp<Relu>>>>&
+        instances);
+
+void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Activation_Mul2_Clamp<PassThrough>>>>&
+        instances);
+
+void add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -89,17 +119,23 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
-                    add_device_conv2d_perchannel_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_perchannel_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_perchannel_quantization_int8_instances(op_ptrs);
+                }
                 else if constexpr(is_same_v<Activation, Relu>)
-                    add_device_conv2d_relu_perchannel_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(op_ptrs);
+                }
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
index 410be4a57bfce65c9c6877ac8508d3cd2921f69f..7c1eb4e429c727099950cb8d722051431c4064bb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -17,13 +17,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
-void add_device_conv2d_perlayer_quantization_int8_instances(
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+void add_device_conv2d_dl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -33,12 +33,42 @@ void add_device_conv2d_perlayer_quantization_int8_instances(
                                                               Activation_Mul_Clamp<PassThrough>>>>&
         instances);
 
-void add_device_conv2d_relu_perlayer_quantization_int8_instances(
+void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Activation_Mul_Clamp<Relu>>>>&
+        instances);
+
+void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Activation_Mul_Clamp<PassThrough>>>>&
+        instances);
+
+void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -87,16 +117,22 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
-                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
-                    add_device_conv2d_perlayer_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_perlayer_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_perlayer_quantization_int8_instances(op_ptrs);
+                }
                 else if constexpr(is_same_v<Activation, Relu>)
-                    add_device_conv2d_relu_perlayer_quantization_int8_instances(op_ptrs);
+                {
+                    add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(op_ptrs);
+                    add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(op_ptrs);
+                }
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
index 550a7b034509af43412c0faf5ac2fb52fb8b2eb3..9930b1a6fa04a5225432a6f8b0499293189b66be 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
index 90cfe837df62e145105f964967e5180ce6b2e768..c9c1475f182578f83288a3b0c1e9db942af494a7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -76,8 +76,16 @@ template <typename InDataType,
           bool PropagateNan,
           bool OutputIndex>
 void add_device_reduce_instance_blockwise(
-    std::vector<DeviceReducePtr<Rank, NumReduceDim, InElementwiseOp, AccElementwiseOp>>&
-        device_op_instances)
+    std::vector<DeviceReducePtr<InDataType,
+                                AccDataType,
+                                OutDataType,
+                                Rank,
+                                NumReduceDim,
+                                ReduceOperation,
+                                InElementwiseOp,
+                                AccElementwiseOp,
+                                PropagateNan,
+                                OutputIndex>>& device_op_instances)
 {
     static_for<0, std::tuple_size<reduce_configuration_1_instances_blockwise>::value, 1>{}(
         [&](auto i) {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp
index 521d93e6001dbf2662ec52c2955416d94f6b10e5..4dd5569cee34aef47aca01c967ba80aca4b9ec7c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp
index fe3fd6c0a7b77620654577dbfa4774ffabe01cb0..d52310a3f3b1b8aecfcb0bfc97f9641943cd9d55 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp
index 52a2b69cdd2175713f5e08a1cbbfe6f2da59360f..025500764ea6730b784cbafec1083cbc2cc0af51 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp
index ee4fee41ea4c38730f9c494ccc3774672856e8cc..2314e94980fb927b034249ecbafa7877056e6cae 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp
index 3abdb7f9588bd11686146980bfc3c1bd333a5aed..5c2bff16cccf6377d347992ab1224bcf27ecff0a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp
index b0dbcf31dd865529e702333188af0bbfa68d9bfe..a1279eeccbf0e38f3862b54aad87204ec44bf2da 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp
index 7bbf3df0a375af62c1ecdbbc47e13fb0e6779e78..aff6a2542bf7a195b090c3a3ee8d1ff4c42f2776 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp
index 559f322261e7a7ffcb8473058a02623b0dedf197..be8da2243fae00df87d2d46d34f1d709dfa32e46 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp
index 28c96107893f5be14474f883a455c601000fd418..652984ae22171b9e0b1941d8379df994cbc30a9c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp
index 5080d2863644f37b71238ca78198c3150575aa71..be60d1b32db50457684a8059f25cc29dc34f6467 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp
index 0d24d15371d02e7f6492f93ee3da7f7ada6073d6..27e3aa53e38911afaf2d47c6900a1fb09ab4ec89 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp
index c806e807c8ed400c59460c72666c4e9277b0ded9..f7f4870a97e98b473773ec70e167bc4818ab469a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp
index b7c046e751f8816b3ccf1bedf806391b71afc491..790b5a92ad7532a26d4663b37189e6d9346fea27 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp
index 771bec1c95b9ba4e47c5741dafe42104a5abcaa3..ec3bc852e8e97e517aa6f18be36b3a40646fa542 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp
index c1fe8addba117e9b92de6fad69220263e8176328..8c0c065670273a5a7ade3ef4f234236589f8ae3e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp
index 6bc0662fea97b0221e20201a178423a6524b9935..8631495df12d6cf9a9154b2a976c80a719352d48 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp
index 6f8005132decf43667334cf1c39f38c8b306af35..59849c2d459b6fd8fbd81adad00fac4493e19686 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp
index c771ac4fab9beeee4a1b3516d1fcfd0fd62bd2b2..33cbea85efefe3d7509888dd3c15d07dca28dc5c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp
index b9ddbb9aea2e14043a948a3b322b8a7cf7330ecc..386182957514384e6c68fbacc961ade931817392 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp
index 390a719ceb1951369b8cd628ced732aed640b87b..f7c05b20a7ce8ff8acf13de3b33b2f52033ce2cb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp
index 2a9ddbc61b3c21713bc50d370ff040efea46435d..2dfecc6b07b2819efd80135a0269d820bb2abaa4 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp
index 5746884442880571f8e00f49e5d4144adb951c60..a687938965e34a45c05d356e94983494fb8b6570 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp
index ad0f2357e05728179c8c75d22808a125a3cf9340..cf9c268a6476be9b62e76742a7d35293dffcbf0d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp
index c7d95276380553f4c3da54813329097a901b13c0..aeb578f948cf2e3d809e44d2975baceb7d20112f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp
index ec56229937a850d2f2708780decc2e024eb8e9f0..73480262c66e99af5afd1d360189585f7fff7410 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp
index 48f66da659b1e39bf64344a1a9339b034207a3ff..74293553e1a168b0a75592bc4f77d9325cfafe60 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp
index fabfa5b4c6f7a335dd007ad95137bae5de876b8a..8a91f76b34faf0b52378ebdb3462d85c20631a12 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp
index e08faec2000ee51745131297fcb4d75ea29c46e4..0ff2c30f346047d092d0ecd9b2e952fd99fe78d6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp
index a1e692aae38ade1a30b96873340aff721e11984b..932299008eba691e5b2f704ca47a00b9159ab5ae 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp
index e9654e8cceba39d9115d2cf2222dc3d72b75d49b..c902e80a2c731f76f4608d4501ad711a1f1a68be 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp
index 78244213097ba7c11168b2f0a7c949133daaac3f..e7a1369277cddb95cd9da2d06a2b16fdad643423 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp
index df323d40b39649599807bf1e303135d0cbe551cf..2d1ab69fc75f74ae9128824f842b1e7f89501db3 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp
index 8c08e5ef2f0365e36125d141f95474573dce5bf6..9f782c11b6dd306c14e1027ff083baa17821b5f5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
index acf55d068392c191912c3f04c167ac185997c031..bb45295ed82e8ed5d6602e3642002f6783bb31fd 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -76,8 +76,16 @@ template <typename InDataType,
           bool PropagateNan,
           bool OutputIndex>
 void add_device_reduce_instance_multiblock_atomic_add(
-    std::vector<DeviceReducePtr<Rank, NumReduceDim, InElementwiseOp, AccElementwiseOp>>&
-        device_op_instances)
+    std::vector<DeviceReducePtr<InDataType,
+                                AccDataType,
+                                OutDataType,
+                                Rank,
+                                NumReduceDim,
+                                ReduceOperation,
+                                InElementwiseOp,
+                                AccElementwiseOp,
+                                PropagateNan,
+                                OutputIndex>>& device_op_instances)
 {
     static_for<0,
                std::tuple_size<reduce_configuration_1_instances_multiblock_atomic_add>::value,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp
index f5102f49770af8f3e0cb057a74ef31c925469dd0..5814b84bb7b295e261b86817b4698ab01284ad41 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp
index ec513113d9e9ed08806b3560a65e6f1caa656e43..f1a609a6f483d979c394ce5b3bd74b19eb14a803 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp
index 3a3d53b8c67b8416113256466f0e410ae998614a..489bef1f561102eb05e9bfc58da3cf45e6bcf54a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp
index bbf439896430e1b88ab6560b9279086958cb75b1..2507d82308da420c51d3c4b1e2a6071aedf186a0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp
index 55147a60e56422b703ae347e56eafc8a66f644f7..df4b7b7b493516a1e01fdc2be435f255bf5cd024 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp
index 4bff06c6afe30abbec780e147d975fcf7ce2d68f..2748b9b9d0434949c50c1c0e4f2e76f2b25bb8af 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp
index daffa1aa4d43fb5b751931c345938282421e17f4..26230d5456c533175b62e54ead132486c80b494b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 // clang-format on
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp
index 52c4171123f87bd19eed10057f2ef5df07e2b444..d5e1499aefc0ada2270f6efdda5776613f9a9c5f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 // clang-format on
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp
index 2f358b06e0e55fa6df1ad1e842e7548c3f81c081..cbc8befb3d39c36a199159b9a1d3759e17f2ce4a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp
index 84c99dcc5759f0d4101d465e84e1e525ffd8fb7e..574cc3dd3e015b4194cb0f584f61dbc8afb6e4e5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
index dfcc8dd8548624f19e16f1521197018e6e9fb7d7..04bab9d283fa83f57eeae64831966fe115dde9c2 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -62,8 +62,16 @@ template <typename InDataType,
           bool PropagateNan,
           bool OutputIndex>
 void add_device_reduce_instance_threadwise(
-    std::vector<DeviceReducePtr<Rank, NumReduceDim, InElementwiseOp, AccElementwiseOp>>&
-        device_op_instances)
+    std::vector<DeviceReducePtr<InDataType,
+                                AccDataType,
+                                OutDataType,
+                                Rank,
+                                NumReduceDim,
+                                ReduceOperation,
+                                InElementwiseOp,
+                                AccElementwiseOp,
+                                PropagateNan,
+                                OutputIndex>>& device_op_instances)
 {
     using cfg1 = ReductionConfiguration_1<256, 256, 1>;
 
@@ -82,6 +90,7 @@ void add_device_reduce_instance_threadwise(
                                                             AccElementwiseOp,
                                                             PropagateNan,
                                                             OutputIndex,
+                                                            false,
                                                             false, // HaveIndexInputIfOutputIndex
                                                             cfg1::BlockSize_,
                                                             cfg2::MThreadSliceSize_,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp
index 4168508b28dfd3f255696a76e198e5cf6c5692b9..6c50d222ecaf19d1026f7cab1478b413de77af10 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp
index 317006e3a5c7c91543eb0bf81040ce86d1a41dc9..5e91f53b0f99864a874fd81f9889cf721780c980 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp
index fc7718ddc044799dde80dff814b41d2beb37f6c7..171803156051497dce6d05bea27663fdc831dbbb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp
index e6616386ca44280dc5d3c1389acff83e58d97a27..03bff5912dac549457c5c51b1e5b501b9ef18f50 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp
index a9441b8e8eac32e6d96dfb923f87bd68bb32b2d3..676af19336c5ca205ccabf31028bd57dfe577d49 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp
index 6820ace8cf0358bd3d2ecf97c3005699fe9b7d54..c67c0a41f85154a9ef5c230958e0e88893d31915 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp
index ab3d4e6e2c4f3ce4565e6ceec25527587be8b499..c9058a9b1d33a2f11f0da600edddb436a2941305 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp
index ee08c9635b9ee109c19153fc4b2f959731e5380e..1fbd3c6e7de827c71e01a601b7edf724cef9a6f2 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp
index 1007ca27bb976c6ea515b521762a02d385022057..89496a4ed16926eec96ae1b5ea560b3b1fd90637 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp
index 1d562c49991c9b4f34db407a5847ed1b1144a7ff..fb79ce8adf39c6fd732d59c4a19f82d8d253cf07 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp
index 5aac638b1eb444289b926c37a171ac6775ac86f2..7915bd2630a25aab209cb42ba69087b54a1a0d76 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp
index 7a3c7640973756e332b695b090aff45e037a8a9f..d769fe8154918b80a2fcbadea95ffe31c57cb128 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp
index 4685d7b5d551b211f96c4ad9c590b42e77cb8431..49938d4434ca072b789aaeac75a8903d9afed112 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp
index 1de338fb4889dd9c2954b030d8b31866afbd07a6..810249994f00d95cb00352eb28ba464f6ecd7856 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp
index e86c41a949775a3e4cef55c7f0e91484212a7a78..6640ffec700bba666962436a63b62ab828cfc719 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp
index 2ca9008560bbef4fab228000b88f7f059868aecd..441bea275c164c60ff9e73af2a30c3a021be0090 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp
index 38380e71ec0f289da2f0c04fe0db7d028bf4453c..05912e1e960f33cef0d82f56798326e3dc65ea2b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp
index 04c5f3e6585ec2f336b9888405ee4b33657a5aa5..b0e16cfeaaef2866134c1dfc5d2a22908404a499 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp
index fef5d408845cd7a76427ea5142212728345efe6d..0ee0f7d0fe156e6fb9a618e0bc66d2a9f4e2e7c4 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp
index 2416f614c34eae56cedfbe41f3c2d1b8f365ef96..feb9d99eba11da1c6b2cddadffd80d3fdb4cb7c2 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp
index fbd0285ae82a340863fd5c41bc0a1e48a3401dbe..a9f0d77cb885b3be117c0621357d12ecac0c599b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp
index 103b85a011d578477268ad209725b80e92c52e2b..bd1f72250a36137736e7bdd3dc5a11d842103539 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp
index e01f590f0eae7ecf47683d3463fbe03960361fcb..60189fd3047b77a3c4ef4f34be9a75ab69dc73ac 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp
index 14a7459bb8a7b631f92f45792e12880a65a09359..b5aa9ce613228d41a89c2787b8bf275f429fcba0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp
index 7dfd806012012016038b5834e7577cdb77708cff..0e20a17ff98e593f93c91861b47449ad6007a6aa 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp
index 7670a27c844bd2cb6a71bae3d33b5138a2fa6da0..c8c1cd7e802022268d323582b7955a366ee743ef 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp
index 8bb85f377928a8c3d1394f9e27da16b9795dcd85..3e71b44673994b1a0fd460191cc29c36b7ac6b97 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp
index a005ba8d426e7f9447e80c70946e95ba463396d1..c8286e56d19377ca1c4a9b4fa9a714687a0b5fcb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp
index 9e8c07eb4f489afe9dfcc6cba3a70f1a6fd1c7b4..36fbc52f113c43a71d6cc0c647b01a16f2a16c4f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,10 +15,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+extern template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp
index a69f88f5a9c6d8930ada6c95ec0c0b05a3557b83..0900cffa353f778e8ce544ee7efbbc1ec9e8fd0a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp
index 734b31c1e977b7a4a6d80494208d5560be17dcc4..726b56066b000521a697989b4f5eb807a998aeba 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp
index 237bd969668d5a24d58234809c0628b7e7239e2f..0ae9838607d53897431229488e177229d88561b8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -15,14 +15,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+extern template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..a3a39b7ccc7216b371a540cb8eaebee075cbe4d6
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp
@@ -0,0 +1,117 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+#include "ck/tensor_operation/gpu/device/device_reduce.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp"
+#include "ck/utility/reduction_operator.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
+          index_t NumReduceDim,
+          typename ReduceOperation,
+          typename InElementwiseOp,
+          typename AccElementwiseOp,
+          bool PropagateNan,
+          bool OutputIndex>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceReduce<InDataType,
+                                                                                 AccDataType,
+                                                                                 OutDataType,
+                                                                                 Rank,
+                                                                                 NumReduceDim,
+                                                                                 ReduceOperation,
+                                                                                 InElementwiseOp,
+                                                                                 AccElementwiseOp,
+                                                                                 PropagateNan,
+                                                                                 OutputIndex>>
+{
+    using DeviceOp = DeviceReduce<InDataType,
+                                  AccDataType,
+                                  OutDataType,
+                                  Rank,
+                                  NumReduceDim,
+                                  ReduceOperation,
+                                  InElementwiseOp,
+                                  AccElementwiseOp,
+                                  PropagateNan,
+                                  OutputIndex>;
+
+    using DeviceOpPtr = DeviceReducePtr<InDataType,
+                                        AccDataType,
+                                        OutDataType,
+                                        Rank,
+                                        NumReduceDim,
+                                        ReduceOperation,
+                                        InElementwiseOp,
+                                        AccElementwiseOp,
+                                        PropagateNan,
+                                        OutputIndex>;
+
+    static auto GetInstances()
+    {
+        std::vector<DeviceOpPtr> op_ptrs;
+
+        constexpr bool out_support_atomic_add =
+            ck::reduce::InMemoryDataOperationSupportedOnDataType<
+                InMemoryDataOperationEnum::AtomicAdd,
+                OutDataType>::value;
+        constexpr bool op_support_atomic_add =
+            std::is_same<ReduceOperation, ReduceAdd>::value &&
+            (std::is_same<AccElementwiseOp, PassThrough>::value ||
+             std::is_same<AccElementwiseOp, UnaryDivide>::value);
+        constexpr bool use_atomic_add = (out_support_atomic_add && op_support_atomic_add);
+
+        add_device_reduce_instance_threadwise<InDataType,
+                                              AccDataType,
+                                              OutDataType,
+                                              Rank,
+                                              NumReduceDim,
+                                              ReduceOperation,
+                                              InElementwiseOp,
+                                              AccElementwiseOp,
+                                              PropagateNan,
+                                              OutputIndex>(op_ptrs);
+
+        add_device_reduce_instance_blockwise<InDataType,
+                                             AccDataType,
+                                             OutDataType,
+                                             Rank,
+                                             NumReduceDim,
+                                             ReduceOperation,
+                                             InElementwiseOp,
+                                             AccElementwiseOp,
+                                             PropagateNan,
+                                             OutputIndex>(op_ptrs);
+
+        if constexpr(use_atomic_add)
+        {
+            add_device_reduce_instance_multiblock_atomic_add<InDataType,
+                                                             AccDataType,
+                                                             OutDataType,
+                                                             Rank,
+                                                             NumReduceDim,
+                                                             ReduceOperation,
+                                                             InElementwiseOp,
+                                                             AccElementwiseOp,
+                                                             PropagateNan,
+                                                             OutputIndex>(op_ptrs);
+        };
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
index 36eb092f0f0707b266ef9a1d7c3b5248947c1e87..26815f1447ce241bdd53003bb43d3a9df25b684e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -9,34 +9,33 @@
 #include "ck/ck.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 #include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_softmax_f16_f16_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>&);
-void add_device_softmax_f16_f16_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>&);
-
-void add_device_softmax_f32_f32_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>&);
-void add_device_softmax_f32_f32_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>&);
-
-void add_device_softmax_i8_i8_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>&);
-void add_device_softmax_i8_i8_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>&);
-
-template <typename InDataType, typename AccDataType, typename OutDataType, index_t Rank>
-struct DeviceOperationInstanceFactory<
-    ck::tensor_operation::device::
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>>
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
+          index_t NumReduceDim>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceSoftmax<InDataType,
+                                                                                  AccDataType,
+                                                                                  OutDataType,
+                                                                                  PassThrough,
+                                                                                  PassThrough,
+                                                                                  Rank,
+                                                                                  NumReduceDim>>
 {
-    using DeviceOp =
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>;
+    using DeviceOp = DeviceSoftmax<InDataType,
+                                   AccDataType,
+                                   OutDataType,
+                                   PassThrough,
+                                   PassThrough,
+                                   Rank,
+                                   NumReduceDim>;
 
     static auto GetInstances()
     {
@@ -46,25 +45,49 @@ struct DeviceOperationInstanceFactory<
                      std::is_same_v<OutDataType, F16>)
         {
             if constexpr(Rank == 3)
-                add_device_softmax_f16_f16_rank3_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f16_f16_rank3_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f16_f16_rank3_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f16_f16_rank3_reduce3_instances(op_ptrs);
+            }
             else if constexpr(Rank == 4)
-                add_device_softmax_f16_f16_rank4_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f16_f16_rank4_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f16_f16_rank4_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f16_f16_rank4_reduce3_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 4)
+                    add_device_softmax_f16_f16_rank4_reduce4_instances(op_ptrs);
+            }
         }
         else if constexpr(std::is_same_v<InDataType, F32> && std::is_same_v<AccDataType, F32> &&
                           std::is_same_v<OutDataType, F32>)
         {
             if constexpr(Rank == 3)
-                add_device_softmax_f32_f32_rank3_instances(op_ptrs);
-            else if constexpr(Rank == 4)
-                add_device_softmax_f32_f32_rank4_instances(op_ptrs);
-        }
-        else if constexpr(std::is_same_v<InDataType, I8> && std::is_same_v<AccDataType, F32> &&
-                          std::is_same_v<OutDataType, I8>)
-        {
-            if constexpr(Rank == 3)
-                add_device_softmax_i8_i8_rank3_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f32_f32_rank3_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f32_f32_rank3_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f32_f32_rank3_reduce3_instances(op_ptrs);
+            }
             else if constexpr(Rank == 4)
-                add_device_softmax_i8_i8_rank4_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f32_f32_rank4_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f32_f32_rank4_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f32_f32_rank4_reduce3_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 4)
+                    add_device_softmax_f32_f32_rank4_reduce4_instances(op_ptrs);
+            }
         }
 
         return op_ptrs;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
deleted file mode 100644
index 83f52fc3ee7a9547161ce1284c2844976819c0ec..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f16_f16_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
-void add_device_softmax_f16_f16_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
index 046ff5780556331a6afc697f87dbbf02dcf1ff25..3fd2bd089ed8fdffa2d40c7b6b10ae146322ee6b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
index 8e6a226f6a13fb1dcd7d61b56b476f38072b6a98..210fdc0a58548f8388602cb960ae0b3c019e05ba 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
index 518fa5f98679099189f15dfe5a57e3ac0324e3fa..894fb034d0dfa86ec85d76b7f015662af1d775d1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
index 10016cdd707cbcb46282e540e15e6b79a67c6a78..708ef0ce130e1095c4c88b520d1dbdbcabf23685 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
index cdd5a3cd7b6fb6806db3de0a10b7ed87b29c5c2d..6754e5ceffa02d4bcad7919c7ea25b0dd85c0cc4 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
index a8be272e02014495cb42d0509b50075f2d7da707..5e111176e198d77b8e0e426f2a578cd9c3dd2b3f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
index ec8296ff22fb6591152a3dcfad63c250c70cabb8..a3cecb32f83f707f54a53b7ed107f2662f4a2f08 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 4>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
index b3877c4bb3f6217427f69c07f7073751b254c938..8c0782daa556168ac359f68d301080720c622e1b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <tuple>
 
@@ -16,7 +16,6 @@ template <index_t Rank, index_t Reduce>
 using device_softmax_f16_f16_instances = std::tuple<
     // clang-format off
     //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
-    // fallback kernel
     DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               1,              1>,
     DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               8,              8>,
     DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,                8,              1,               8,              8>,
@@ -33,6 +32,13 @@ using device_softmax_f16_f16_instances = std::tuple<
     // clang-format on
     >;
 
+template <index_t Rank, index_t Reduce>
+using device_softmax_f16_f16_generic_instance = std::tuple<
+    // clang-format off
+    DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,        64,                  8,                  8,                1,                1,              1,               1,              1>
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
deleted file mode 100644
index a6d9a359f4622ec9534ab090d46e363bd9101177..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f32_f32_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
-void add_device_softmax_f32_f32_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
index 6621a2c867ac90402a853e5c2065663a96220927..4cc469025335831b8c502e0693e92894ba454202 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
index 3dfac98ed8beeee6060b6ba3f6b9b1ff19f660fe..65724d7888ac9801cc2dcd97ce56dd8681ca6cfe 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
index 6d2a0c932500e874641d8cef9fd65b4d280c89b6..13bd45598eca17c81df5b8eba6fe41f3400057b5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
index 97dd3dcb18aed5c228313a961994096445948bbf..d58b424ee94ffad33440525e190c7adf45598eb2 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
index 58f8760acccd1b571df40ea9a08946c1f5f9764f..378e45eeb783d85a6da303349c7bd7b0329f7fc1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
index df8d31f0da76b0f027b4ee4818006dd331e38454..293df08c7e9e8edd9a419b7e94a5d83fa46efe8f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
index 1bd773227e172278a466d7650e90fd39ec3e56f8..e503a9fec1f5c13e71b297d985c3de8c15497758 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 4>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
index 16f129d2d07c19a926c75f1e878e8db3fb9bf821..90c5ddc8a01b7fd0ef58a8f744133a0635e2fe29 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <tuple>
 
@@ -16,7 +16,7 @@ template <index_t Rank, index_t Reduce>
 using device_softmax_f32_f32_instances = std::tuple<
     // clang-format off
     //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
-    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               1,               1>, // fallback kernel
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               1,               1>,
     DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               4,               4>,
     DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,                8,              1,               4,               4>,
     DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,                8,              1,               4,               4>,
@@ -32,6 +32,13 @@ using device_softmax_f32_f32_instances = std::tuple<
     // clang-format on
     >;
 
+template <index_t Rank, index_t Reduce>
+using device_softmax_f32_f32_generic_instance = std::tuple<
+    // clang-format off
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,        64,                  8,                  8,                1,                1,              1,               1,               1>
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
deleted file mode 100644
index f80f712ff5e501d6516d9c7210a59276a96b6541..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
-void add_device_softmax_i8_i8_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
deleted file mode 100644
index 6f9952e7d58a448cfea0e3c6ba84af24e9de4e52..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
deleted file mode 100644
index 2cbd13a1ba5226316363cfb07dfd23adfac4aa9f..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
deleted file mode 100644
index 7b12522a85987f85e83a6f8afb34b97844507c49..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
deleted file mode 100644
index 54d477f80c5b222cd5a76430dae3bbd4db81068c..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
deleted file mode 100644
index 4ffc44e3a9217e689297bceb81bd6a38c41cd1c3..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
deleted file mode 100644
index 08cbb81272f9ff5e7ce9a9fdd9075d81f1ed7211..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
deleted file mode 100644
index 187d034b95ac506426da4fa1b8240b61e2517e46..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
deleted file mode 100644
index 7fc9ed69198f1e7bf447a6baa9dedb3286440a94..0000000000000000000000000000000000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <tuple>
-
-#include "ck/ck.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp"
-#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-template <index_t Rank, index_t Reduce>
-using device_softmax_i8_i8_instances = std::tuple<
-    // clang-format off
-    //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
-    // fallback kernel
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,               16,              1,               1,              1>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               32,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               64,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               32,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               64,              1,              16,             16>,
-    // Reduction on middle dimensions
-    // InSrcVectorDim is 0 since we want to coalesce reads on M dimension
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                8,                8,              0,               1,              1>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                 32,                  8,               32,                8,              0,              16,              8>
-    // clang-format on
-    >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
index 03be6e2bc7c15ee350569f7ed97ebf2d3ad3b515..10f99acb8d6fe4c5d1fb0a065231995a3d870392 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
@@ -1,8 +1,19 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp"
diff --git a/library/include/ck/library/utility/algorithm.hpp b/library/include/ck/library/utility/algorithm.hpp
index 86f04dd362397e774c9821a16a7c412dea000937..57136f8a2a19fbdaa2e0276699c786c33cc61473 100644
--- a/library/include/ck/library/utility/algorithm.hpp
+++ b/library/include/ck/library/utility/algorithm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/check_err.hpp b/library/include/ck/library/utility/check_err.hpp
index e6d5e893cb89661657d49423e0d4d54c3853ce08..bbd02281ec45130599bb53ad246c919874527327 100644
--- a/library/include/ck/library/utility/check_err.hpp
+++ b/library/include/ck/library/utility/check_err.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/conv_common.hpp b/library/include/ck/library/utility/conv_common.hpp
index 6fad9f7d77d349853e63512ae486a57e7ceb7b6f..085454f42dee006cf7c48ebd18804df92fa36a37 100644
--- a/library/include/ck/library/utility/conv_common.hpp
+++ b/library/include/ck/library/utility/conv_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp b/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp
index 2b4f63b28b8d2f1e59cef67b99a577314faf4ae2..ff697fb71c0340df3f3aac6953ce76cfdef4b278 100644
--- a/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp
+++ b/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/convolution_parameter.hpp b/library/include/ck/library/utility/convolution_parameter.hpp
index f4a2b56f75a6b40a476080e74d386fab524286a4..df6efca10808beb96535ef69a3ba6116a449b9e3 100644
--- a/library/include/ck/library/utility/convolution_parameter.hpp
+++ b/library/include/ck/library/utility/convolution_parameter.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/device_memory.hpp b/library/include/ck/library/utility/device_memory.hpp
index 3c4ece44068dd2dafe5869f6943a0b27b2508355..1c16ff59165af029e2f81d8364daf07f844daaac 100644
--- a/library/include/ck/library/utility/device_memory.hpp
+++ b/library/include/ck/library/utility/device_memory.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -14,6 +14,10 @@ __global__ void set_buffer_value(T* p, T x, uint64_t buffer_element_size)
     }
 }
 
+/**
+ * @brief Container for storing data in GPU device memory
+ *
+ */
 struct DeviceMem
 {
     DeviceMem() = delete;
diff --git a/library/include/ck/library/utility/fill.hpp b/library/include/ck/library/utility/fill.hpp
index 54d58f362ccaffad209df586e2b9a399120d1d88..c01e139ea0b8f160f790bde4270640c261affa6e 100644
--- a/library/include/ck/library/utility/fill.hpp
+++ b/library/include/ck/library/utility/fill.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -100,6 +100,15 @@ struct FillMonotonicSeq
             return tmp;
         });
     }
+
+    template <typename ForwardRange>
+    auto operator()(ForwardRange&& range) const -> std::void_t<decltype(
+        std::declval<const FillMonotonicSeq&>()(std::begin(std::forward<ForwardRange>(range)),
+                                                std::end(std::forward<ForwardRange>(range))))>
+    {
+        (*this)(std::begin(std::forward<ForwardRange>(range)),
+                std::end(std::forward<ForwardRange>(range)));
+    }
 };
 
 template <typename T>
@@ -112,6 +121,15 @@ struct FillConstant
     {
         std::fill(first, last, value_);
     }
+
+    template <typename ForwardRange>
+    auto operator()(ForwardRange&& range) const -> std::void_t<
+        decltype(std::declval<const FillConstant&>()(std::begin(std::forward<ForwardRange>(range)),
+                                                     std::end(std::forward<ForwardRange>(range))))>
+    {
+        (*this)(std::begin(std::forward<ForwardRange>(range)),
+                std::end(std::forward<ForwardRange>(range)));
+    }
 };
 
 } // namespace utils
diff --git a/library/include/ck/library/utility/host_common_util.hpp b/library/include/ck/library/utility/host_common_util.hpp
index 6f4466e8da0e2b77c1ad981cfb6f730a74a92d71..20a8f234dbbf9f30582ed371c5f1a90cce2b3196 100644
--- a/library/include/ck/library/utility/host_common_util.hpp
+++ b/library/include/ck/library/utility/host_common_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/host_conv.hpp b/library/include/ck/library/utility/host_conv.hpp
deleted file mode 100644
index 8348a3089f4606563caa5582ac427154308e7fa5..0000000000000000000000000000000000000000
--- a/library/include/ck/library/utility/host_conv.hpp
+++ /dev/null
@@ -1,152 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-#include "host_tensor.hpp"
-#include "conv_common.hpp"
-
-template <typename TIn,
-          typename TWei,
-          typename TOut,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-void host_conv_nchw_kcyx_nkhw(const Tensor<TIn>& in,
-                              const Tensor<TWei>& wei,
-                              Tensor<TOut>& out,
-                              const ConvStrides& conv_strides,
-                              const ConvDilations& conv_dilations,
-                              const InLeftPads& in_left_pads,
-                              const InRightPads&)
-{
-    constexpr auto I0 = ck::Number<0>{};
-    constexpr auto I1 = ck::Number<1>{};
-
-    auto f_nchw = [&](auto n, auto k, auto ho, auto wo) {
-        float v = 0;
-        for(int c = 0; c < wei.mDesc.GetLengths()[1]; ++c)
-        {
-            for(int y = 0; y < wei.mDesc.GetLengths()[2]; ++y)
-            {
-                int hi = ho * conv_strides[I0] + y * conv_dilations[I0] - in_left_pads[I0];
-                for(int x = 0; x < wei.mDesc.GetLengths()[3]; ++x)
-                {
-                    int wi = wo * conv_strides[I1] + x * conv_dilations[I1] - in_left_pads[I1];
-                    if(hi >= 0 && hi < in.mDesc.GetLengths()[2] && wi >= 0 &&
-                       wi < in.mDesc.GetLengths()[3])
-                    {
-                        v += ck::type_convert<float>(in(n, c, hi, wi)) *
-                             ck::type_convert<float>(wei(k, c, y, x));
-                    }
-                }
-            }
-        }
-        out(n, k, ho, wo) = ck::type_convert<TOut>(v);
-    };
-
-    make_ParallelTensorFunctor(f_nchw,
-                               out.mDesc.GetLengths()[0],
-                               out.mDesc.GetLengths()[1],
-                               out.mDesc.GetLengths()[2],
-                               out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-}
-
-template <typename TIn,
-          typename TWei,
-          typename TOut,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-void host_conv3d_ndhwc_kzyxc_ndhwk(const Tensor<TIn>& in,
-                                   const Tensor<TWei>& wei,
-                                   Tensor<TOut>& out,
-                                   const ConvStrides& conv_strides,
-                                   const ConvDilations& conv_dilations,
-                                   const InLeftPads& in_left_pads,
-                                   const InRightPads&)
-{
-    using namespace ck;
-
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    const auto Di     = in.mDesc.GetLengths()[1];
-    const auto Hi     = in.mDesc.GetLengths()[2];
-    const auto Wi     = in.mDesc.GetLengths()[3];
-    const auto Z      = wei.mDesc.GetLengths()[1];
-    const auto Y      = wei.mDesc.GetLengths()[2];
-    const auto X      = wei.mDesc.GetLengths()[3];
-    const auto C      = wei.mDesc.GetLengths()[4];
-
-    auto f_ndhwc = [&](auto n, auto do_tmp, auto ho_tmp, auto wo_tmp, auto k) {
-        // do__ must be converted to signed integer, otherwise zmin might be wrong in cases
-        // negative values.
-        const int do_ = static_cast<int>(do_tmp);
-        const int ho  = static_cast<int>(ho_tmp);
-        const int wo  = static_cast<int>(wo_tmp);
-        const int zmin =
-            std::max(0,
-                     (in_left_pads[I0] - do_ * conv_strides[I0] + conv_dilations[I0] - 1) /
-                         conv_dilations[I0]);
-        const int ymin =
-            std::max(0,
-                     (in_left_pads[I1] - ho * conv_strides[I1] + conv_dilations[I1] - 1) /
-                         conv_dilations[I1]);
-        const int xmin =
-            std::max(0,
-                     (in_left_pads[I2] - wo * conv_strides[I2] + conv_dilations[I2] - 1) /
-                         conv_dilations[I2]);
-        const int zmax =
-            std::min(Z, (in_left_pads[I0] - do_ * conv_strides[I0] + Di) / conv_dilations[I0]);
-        const int ymax =
-            std::min(Y, (in_left_pads[I1] - ho * conv_strides[I1] + Hi) / conv_dilations[I1]);
-        const int xmax =
-            std::min(X, (in_left_pads[I2] - wo * conv_strides[I2] + Wi) / conv_dilations[I2]);
-        const int di_min = do_ * conv_strides[I0] + zmin * conv_dilations[I0] - in_left_pads[I0];
-        const int hi_min = ho * conv_strides[I1] + ymin * conv_dilations[I1] - in_left_pads[I1];
-        const int wi_min = wo * conv_strides[I2] + xmin * conv_dilations[I2] - in_left_pads[I2];
-
-        double v = 0;
-
-        const TIn* in_n   = in.mData.data() + n * Di * Hi * Wi * C;
-        const TWei* wei_k = wei.mData.data() + k * Z * Y * X * C;
-
-        int di = di_min;
-        for(int z = zmin; z < zmax; ++z, di += conv_dilations[I0])
-        {
-            const TIn* in_n_di  = in_n + di * Hi * Wi * C;
-            const TWei* wei_k_z = wei_k + z * Y * X * C;
-            int hi              = hi_min;
-
-            for(int y = ymin; y < ymax; ++y, hi += conv_dilations[I1])
-            {
-                const TIn* in_n_di_hi = in_n_di + hi * Wi * C;
-                const TWei* wei_k_z_y = wei_k_z + y * X * C;
-                int wi                = wi_min;
-
-                for(int x = xmin; x < xmax; ++x, wi += conv_dilations[I2])
-                {
-                    const TIn* in_n_di_hi_wi = in_n_di_hi + wi * C;
-                    const TWei* wei_k_z_y_x  = wei_k_z_y + x * C;
-
-                    for(int c = 0; c < C; ++c)
-                    {
-                        v += static_cast<const double>(in_n_di_hi_wi[c]) *
-                             static_cast<const double>(wei_k_z_y_x[c]);
-                    }
-                }
-            }
-        }
-
-        out(n, do_, ho, wo, k) = v;
-    };
-
-    make_ParallelTensorFunctor(f_ndhwc,
-                               out.mDesc.GetLengths()[0],
-                               out.mDesc.GetLengths()[1],
-                               out.mDesc.GetLengths()[2],
-                               out.mDesc.GetLengths()[3],
-                               out.mDesc.GetLengths()[4])(std::thread::hardware_concurrency() - 4);
-}
diff --git a/library/include/ck/library/utility/host_gemm.hpp b/library/include/ck/library/utility/host_gemm.hpp
index 44036d0234375824ed3aa773ffe71de02b84c570..5eb7e3b8c9745c48390482622d361a2ddd491159 100644
--- a/library/include/ck/library/utility/host_gemm.hpp
+++ b/library/include/ck/library/utility/host_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/host_reduction.hpp b/library/include/ck/library/utility/host_reduction.hpp
deleted file mode 100644
index 7c0c969ac592a459c5df37fbfcf807f2f500df37..0000000000000000000000000000000000000000
--- a/library/include/ck/library/utility/host_reduction.hpp
+++ /dev/null
@@ -1,374 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-#include <array>
-#include <functional>
-
-#include "ck/utility/data_type.hpp"
-#include "ck/utility/reduction_enums.hpp"
-#include "ck/utility/reduction_common.hpp"
-#include "ck/utility/reduction_functions_accumulate.hpp"
-#include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-
-template <int NDim>
-static void get_all_indexes(const std::array<size_t, NDim>& dimLengths,
-                            std::vector<std::array<size_t, NDim>>& indexes)
-{
-    static_assert(NDim >= 1, "NDim >= 1 is required to use this function!");
-
-    if constexpr(NDim == 1)
-    {
-        for(size_t i = 0; i < dimLengths[0]; i++)
-        {
-            std::array<size_t, 1> index{i};
-
-            indexes.push_back(index);
-        };
-    }
-    else
-    {
-        std::array<size_t, NDim - 1> partial_dim_lengths;
-
-        for(int i = 0; i < NDim - 1; i++)
-            partial_dim_lengths[i] = dimLengths[i + 1];
-
-        std::vector<std::array<size_t, NDim - 1>> partial_indexes;
-
-        get_all_indexes<NDim - 1>(partial_dim_lengths, partial_indexes);
-
-        for(size_t i = 0; i < dimLengths[0]; i++)
-            for(const auto& index : partial_indexes)
-            {
-                std::array<size_t, NDim> extIndex;
-
-                extIndex[0] = i;
-
-                for(int k = 0; k < NDim - 1; k++)
-                    extIndex[k + 1] = index[k];
-
-                indexes.push_back(extIndex);
-            };
-    };
-};
-
-template <int NDim>
-static size_t get_offset_from_index(const std::array<size_t, NDim>& strides,
-                                    const std::array<size_t, NDim>& index)
-{
-    size_t offset = 0;
-
-    for(int i = 0; i < NDim; i++)
-        offset += strides[i] * index[i];
-
-    return (offset);
-};
-
-template <int NDim>
-static size_t get_offset_from_index(const std::vector<size_t>& strides,
-                                    const std::array<size_t, NDim>& index)
-{
-    size_t offset = 0;
-
-    for(int i = 0; i < NDim; i++)
-        offset += strides[i] * index[i];
-
-    return (offset);
-};
-
-template <typename InDataType,
-          typename AccDataType,
-          typename OutDataType,
-          typename ReduceOperation,
-          typename InElementwiseOperation,
-          typename AccElementwiseOperation,
-          int Rank,
-          int NumReduceDim,
-          bool PropagateNan,
-          bool OutputIndex>
-struct ReductionHost
-{
-    using IndexDataType = int32_t;
-
-    static constexpr int NumInvariantDim = Rank - NumReduceDim;
-
-    std::vector<size_t> outStrides;
-
-    IndexDataType divider;
-
-    std::array<size_t, NumReduceDim> reduceLengths;
-    std::array<size_t, NumReduceDim> reduceStrides;
-    std::array<size_t, NumInvariantDim> invariantLengths;
-    std::array<size_t, NumInvariantDim> invariantStrides;
-
-    std::vector<std::array<size_t, NumReduceDim>> reduce_dim_indexes;
-    std::vector<std::array<size_t, NumInvariantDim>> invariant_dim_indexes;
-
-    ReductionHost(HostTensorDescriptor& inDesc,
-                  HostTensorDescriptor& outDesc,
-                  const std::array<int, NumInvariantDim> invariantDims,
-                  const std::array<int, NumReduceDim> reduceDims)
-    {
-        // this->outLengths = to_int_vector(outDesc.GetLengths());
-        this->outStrides = outDesc.GetStrides();
-
-        int product = 1;
-
-        for(int i = 0; i < NumReduceDim; i++)
-        {
-            reduceLengths[i] = inDesc.GetLengths()[reduceDims[i]];
-            reduceStrides[i] = inDesc.GetStrides()[reduceDims[i]];
-            product *= inDesc.GetLengths()[reduceDims[i]];
-        };
-
-        divider = product;
-
-        for(int i = 0; i < NumInvariantDim; i++)
-        {
-            invariantLengths[i] = inDesc.GetLengths()[invariantDims[i]];
-            invariantStrides[i] = inDesc.GetStrides()[invariantDims[i]];
-        };
-
-        reduce_dim_indexes.clear();
-        get_all_indexes<NumReduceDim>(reduceLengths, reduce_dim_indexes);
-
-        if constexpr(NumInvariantDim > 0)
-        {
-            invariant_dim_indexes.clear();
-            get_all_indexes<NumInvariantDim>(invariantLengths, invariant_dim_indexes);
-        };
-    };
-
-    void Run(float alpha,
-             const InDataType* in_data,
-             float beta,
-             OutDataType* out_data,
-             IndexDataType* out_indices,
-             InElementwiseOperation in_elementwise_op,
-             AccElementwiseOperation acc_elementwise_op)
-    {
-        if constexpr(OutputIndex)
-        {
-            RunImpl_with_index(
-                alpha, in_data, beta, out_data, out_indices, in_elementwise_op, acc_elementwise_op);
-        }
-        else
-        {
-            RunImpl_no_index(alpha, in_data, beta, out_data, in_elementwise_op, acc_elementwise_op);
-        };
-    };
-
-    void RunImpl_with_index(float alpha,
-                            const InDataType* in_data,
-                            float beta,
-                            OutDataType* out_data,
-                            IndexDataType* out_indices,
-                            InElementwiseOperation in_elementwise_op,
-                            AccElementwiseOperation acc_elementwise_op)
-    {
-        using ck::float_equal_one;
-        using ck::float_equal_zero;
-        using ck::type_convert;
-
-        using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
-                                                                        ReduceOperation,
-                                                                        AccDataType,
-                                                                        IndexDataType>;
-
-        if constexpr(NumInvariantDim == 0)
-        {
-            AccDataType accuVal     = ReduceOperation::template GetIdentityValue<AccDataType>();
-            IndexDataType accuIndex = 0;
-
-            for(std::size_t i = 0; i < reduce_dim_indexes.size(); i++)
-            {
-                auto offset_reduce =
-                    get_offset_from_index<NumReduceDim>(reduceStrides, reduce_dim_indexes[i]);
-
-                auto currVal = type_convert<AccDataType>(in_data[offset_reduce]);
-
-                in_elementwise_op(currVal, currVal);
-
-                auto currIndex = static_cast<IndexDataType>(i);
-
-                Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
-            };
-
-            acc_elementwise_op(accuVal, accuVal);
-
-            if(!float_equal_one{}(alpha))
-                accuVal *= type_convert<AccDataType>(alpha);
-
-            if(!float_equal_zero{}(beta))
-                accuVal += type_convert<AccDataType>(out_data[0]) * type_convert<AccDataType>(beta);
-
-            out_data[0]    = type_convert<OutDataType>(accuVal);
-            out_indices[0] = accuIndex;
-        }
-        else
-        {
-            auto thread_reduce_func = [&](auto invariant_index) {
-                AccDataType accuVal     = ReduceOperation::template GetIdentityValue<AccDataType>();
-                IndexDataType accuIndex = 0;
-
-                auto offset_invariant =
-                    get_offset_from_index<NumInvariantDim>(invariantStrides, invariant_index);
-
-                for(std::size_t i = 0; i < reduce_dim_indexes.size(); i++)
-                {
-                    auto offset_reduce =
-                        get_offset_from_index<NumReduceDim>(reduceStrides, reduce_dim_indexes[i]);
-
-                    auto currVal =
-                        type_convert<AccDataType>(in_data[offset_invariant + offset_reduce]);
-
-                    in_elementwise_op(currVal, currVal);
-
-                    auto currIndex = static_cast<IndexDataType>(i);
-
-                    Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
-                };
-
-                acc_elementwise_op(accuVal, accuVal);
-
-                if(!float_equal_one{}(alpha))
-                    accuVal *= type_convert<AccDataType>(alpha);
-
-                auto dst_offset =
-                    get_offset_from_index<NumInvariantDim>(outStrides, invariant_index);
-
-                if(!float_equal_zero{}(beta))
-                    accuVal += type_convert<AccDataType>(out_data[dst_offset]) *
-                               type_convert<AccDataType>(beta);
-
-                out_data[dst_offset]    = type_convert<OutDataType>(accuVal);
-                out_indices[dst_offset] = accuIndex;
-            };
-
-            std::size_t num_thread = 1;
-            std::size_t work_per_thread =
-                (invariant_dim_indexes.size() + num_thread - 1) / num_thread;
-
-            std::vector<joinable_thread> threads(num_thread);
-
-            for(std::size_t it = 0; it < num_thread; ++it)
-            {
-                std::size_t iw_begin = it * work_per_thread;
-                std::size_t iw_end =
-                    std::min((it + 1) * work_per_thread, invariant_dim_indexes.size());
-
-                auto f = [=] {
-                    for(std::size_t iw = iw_begin; iw < iw_end; ++iw)
-                    {
-                        thread_reduce_func(invariant_dim_indexes[iw]);
-                    }
-                };
-
-                threads[it] = joinable_thread(f);
-            }
-        };
-    };
-
-    void RunImpl_no_index(float alpha,
-                          const InDataType* in_data,
-                          float beta,
-                          OutDataType* out_data,
-                          InElementwiseOperation in_elementwise_op,
-                          AccElementwiseOperation acc_elementwise_op)
-    {
-        using ck::float_equal_one;
-        using ck::float_equal_zero;
-        using ck::type_convert;
-
-        using Accumulation =
-            ck::detail::AccumulateWithNanCheck<PropagateNan, ReduceOperation, AccDataType>;
-
-        if constexpr(NumInvariantDim == 0)
-        {
-            AccDataType accuVal = ReduceOperation::template GetIdentityValue<AccDataType>();
-
-            for(const auto& reduce_index : reduce_dim_indexes)
-            {
-                auto offset_reduce =
-                    get_offset_from_index<NumReduceDim>(reduceStrides, reduce_index);
-
-                auto currVal = type_convert<AccDataType>(in_data[offset_reduce]);
-
-                in_elementwise_op(currVal, currVal);
-
-                Accumulation::Calculate(accuVal, currVal);
-            };
-
-            acc_elementwise_op(accuVal, accuVal);
-
-            if(!float_equal_one{}(alpha))
-                accuVal *= type_convert<AccDataType>(alpha);
-
-            if(!float_equal_zero{}(beta))
-                accuVal += type_convert<AccDataType>(out_data[0]) * type_convert<AccDataType>(beta);
-
-            out_data[0] = type_convert<OutDataType>(accuVal);
-        }
-        else
-        {
-            auto thread_reduce_func = [&](auto invariant_index) {
-                AccDataType accuVal = ReduceOperation::template GetIdentityValue<AccDataType>();
-
-                auto offset_invariant =
-                    get_offset_from_index<NumInvariantDim>(invariantStrides, invariant_index);
-
-                for(const auto& reduce_index : reduce_dim_indexes)
-                {
-                    auto offset_reduce =
-                        get_offset_from_index<NumReduceDim>(reduceStrides, reduce_index);
-
-                    auto currVal =
-                        type_convert<AccDataType>(in_data[offset_invariant + offset_reduce]);
-
-                    in_elementwise_op(currVal, currVal);
-
-                    Accumulation::Calculate(accuVal, currVal);
-                };
-
-                acc_elementwise_op(accuVal, accuVal);
-
-                if(!float_equal_one{}(alpha))
-                    accuVal *= type_convert<AccDataType>(alpha);
-
-                auto dst_offset =
-                    get_offset_from_index<NumInvariantDim>(outStrides, invariant_index);
-
-                if(!float_equal_zero{}(beta))
-                    accuVal += type_convert<AccDataType>(out_data[dst_offset]) *
-                               type_convert<AccDataType>(beta);
-
-                out_data[dst_offset] = type_convert<OutDataType>(accuVal);
-            };
-
-            std::size_t num_thread = 1;
-            std::size_t work_per_thread =
-                (invariant_dim_indexes.size() + num_thread - 1) / num_thread;
-
-            std::vector<joinable_thread> threads(num_thread);
-
-            for(std::size_t it = 0; it < num_thread; ++it)
-            {
-                std::size_t iw_begin = it * work_per_thread;
-                std::size_t iw_end =
-                    std::min((it + 1) * work_per_thread, invariant_dim_indexes.size());
-
-                auto f = [=] {
-                    for(std::size_t iw = iw_begin; iw < iw_end; ++iw)
-                    {
-                        thread_reduce_func(invariant_dim_indexes[iw]);
-                    }
-                };
-
-                threads[it] = joinable_thread(f);
-            }
-        };
-    };
-};
diff --git a/library/include/ck/library/utility/host_tensor.hpp b/library/include/ck/library/utility/host_tensor.hpp
index 2fce7dd8de57ccc4e05d4575f661bde667ec3258..5812f19d13dbe37f502b8fb4848c78a59cdaaabe 100644
--- a/library/include/ck/library/utility/host_tensor.hpp
+++ b/library/include/ck/library/utility/host_tensor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -13,6 +13,7 @@
 
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/span.hpp"
+#include "ck/utility/type_convert.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/ranges.hpp"
@@ -396,7 +397,7 @@ struct Tensor
         }
         case 6: {
             auto f = [&](auto i0, auto i1, auto i2, auto i3, auto i4, auto i5) {
-                (*this)(i0, i1, i2, i3, i4) = g(i0, i1, i2, i3, i4, i5);
+                (*this)(i0, i1, i2, i3, i4, i5) = g(i0, i1, i2, i3, i4, i5);
             };
             make_ParallelTensorFunctor(f,
                                        mDesc.GetLengths()[0],
@@ -411,6 +412,12 @@ struct Tensor
         }
     }
 
+    template <typename... Is>
+    std::size_t GetOffsetFromMultiIndex(Is... is) const
+    {
+        return mDesc.GetOffsetFromMultiIndex(is...);
+    }
+
     template <typename... Is>
     T& operator()(Is... is)
     {
diff --git a/library/include/ck/library/utility/host_tensor_generator.hpp b/library/include/ck/library/utility/host_tensor_generator.hpp
index 4259862e65e331b30377997c512459e2a21e9383..31ff13aec254a166c6439fdef3fd17907017e147 100644
--- a/library/include/ck/library/utility/host_tensor_generator.hpp
+++ b/library/include/ck/library/utility/host_tensor_generator.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/iterator.hpp b/library/include/ck/library/utility/iterator.hpp
index 9fdc88ea76825e2a1cdf659f95fa2928038867aa..b44e2d8e3c4a6f8b4423b03699539ade2fbfb0d7 100644
--- a/library/include/ck/library/utility/iterator.hpp
+++ b/library/include/ck/library/utility/iterator.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/literals.hpp b/library/include/ck/library/utility/literals.hpp
index a73a2ea054150665557b8e1da473f2cb9877526b..a8bd6303f12f0a36bafb36c1d7f1453eab13392b 100644
--- a/library/include/ck/library/utility/literals.hpp
+++ b/library/include/ck/library/utility/literals.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/numeric.hpp b/library/include/ck/library/utility/numeric.hpp
index 70a7e87ab1c8d94569e705b3f6832c78bac05853..9ee118d4757109416bac18cb54229e93de59b75c 100644
--- a/library/include/ck/library/utility/numeric.hpp
+++ b/library/include/ck/library/utility/numeric.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/op_instance_engine.hpp b/library/include/ck/library/utility/op_instance_engine.hpp
deleted file mode 100644
index 78812e8c81d600ab1258400205b98994ab120d3a..0000000000000000000000000000000000000000
--- a/library/include/ck/library/utility/op_instance_engine.hpp
+++ /dev/null
@@ -1,249 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <cstdlib>
-#include <iostream>
-#include <limits>
-#include <memory>
-#include <stdexcept>
-#include <tuple>
-#include <utility>
-#include <vector>
-
-#include "ck/utility/functional2.hpp"
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace utils {
-
-struct ProfileBestConfig
-{
-    std::string best_op_name;
-    float best_avg_time   = std::numeric_limits<float>::max();
-    float best_tflops     = std::numeric_limits<float>::max();
-    float best_gb_per_sec = std::numeric_limits<float>::max();
-};
-
-/**
- * @brief      This class describes an operation instance(s).
- *
- *             Op instance defines a particular specializations of operator
- *             template. Thanks to this specific input/output data types, data
- *             layouts and modifying elementwise operations it is able to create
- *             it's input/output tensors, provide pointers to instances which
- *             can execute it and all operation specific parameters.
- */
-template <typename OutDataType, typename... InArgTypes>
-class OpInstance
-{
-    public:
-    template <typename T>
-    using TensorPtr      = std::unique_ptr<Tensor<T>>;
-    using InTensorsTuple = std::tuple<TensorPtr<InArgTypes>...>;
-    using DeviceMemPtr   = std::unique_ptr<DeviceMem>;
-    using DeviceBuffers  = std::vector<DeviceMemPtr>;
-
-    OpInstance()                  = default;
-    OpInstance(const OpInstance&) = default;
-    OpInstance& operator=(const OpInstance&) = default;
-    virtual ~OpInstance(){};
-
-    virtual InTensorsTuple GetInputTensors() const         = 0;
-    virtual TensorPtr<OutDataType> GetOutputTensor() const = 0;
-    virtual std::unique_ptr<tensor_operation::device::BaseInvoker>
-    MakeInvokerPointer(tensor_operation::device::BaseOperator*) const = 0;
-    virtual std::unique_ptr<tensor_operation::device::BaseArgument>
-    MakeArgumentPointer(tensor_operation::device::BaseOperator*,
-                        const DeviceBuffers&,
-                        const DeviceMemPtr&) const = 0;
-    virtual std::size_t GetFlops() const           = 0;
-    virtual std::size_t GetBtype() const           = 0;
-};
-
-/**
- * @brief      A generic operation instance run engine.
- */
-template <typename OutDataType, typename... InArgTypes>
-class OpInstanceRunEngine
-{
-    public:
-    using OpInstanceT = OpInstance<InArgTypes..., OutDataType>;
-    template <typename T>
-    using TensorPtr        = std::unique_ptr<Tensor<T>>;
-    using DeviceMemPtr     = std::unique_ptr<DeviceMem>;
-    using InTensorsTuple   = std::tuple<TensorPtr<InArgTypes>...>;
-    using DeviceBuffers    = std::vector<DeviceMemPtr>;
-    using InArgsTypesTuple = std::tuple<InArgTypes...>;
-
-    OpInstanceRunEngine() = delete;
-
-    template <typename ReferenceOp = std::function<void()>>
-    OpInstanceRunEngine(const OpInstanceT& op_instance,
-                        const ReferenceOp& reference_op = ReferenceOp{},
-                        bool do_verification            = true)
-        : op_instance_{op_instance}
-    {
-        in_tensors_ = op_instance_.GetInputTensors();
-        out_tensor_ = op_instance_.GetOutputTensor();
-
-        if constexpr(std::is_invocable_v<ReferenceOp,
-                                         const Tensor<InArgTypes>&...,
-                                         Tensor<OutDataType>&>)
-        {
-            if(do_verification)
-            {
-                ref_output_ = op_instance_.GetOutputTensor();
-                CallRefOpUnpackArgs(reference_op, std::make_index_sequence<kNInArgs_>{});
-            }
-        }
-        AllocateDeviceInputTensors(std::make_index_sequence<kNInArgs_>{});
-        out_device_buffer_ = std::make_unique<DeviceMem>(sizeof(OutDataType) *
-                                                         out_tensor_->mDesc.GetElementSpaceSize());
-        out_device_buffer_->SetZero();
-    }
-
-    virtual ~OpInstanceRunEngine(){};
-
-    template <typename OpInstancePtr>
-    bool Test(const std::vector<OpInstancePtr>& op_ptrs)
-    {
-        bool res{true};
-        for(auto& op_ptr : op_ptrs)
-        {
-            auto invoker  = op_instance_.MakeInvokerPointer(op_ptr.get());
-            auto argument = op_instance_.MakeArgumentPointer(
-                op_ptr.get(), in_device_buffers_, out_device_buffer_);
-            if(op_ptr->IsSupportedArgument(argument.get()))
-            {
-                std::cout << "Testing instance: " << op_ptr->GetTypeString() << std::endl;
-                invoker->Run(argument.get());
-                out_device_buffer_->FromDevice(out_tensor_->mData.data());
-                if(!ref_output_)
-                {
-                    throw std::runtime_error(
-                        "OpInstanceRunEngine::Test: Reference value not availabe."
-                        " You have to provide reference function.");
-                }
-                // TODO: enable flexible use of custom check_error functions
-                bool inst_res = CheckErr(out_tensor_->mData, ref_output_->mData);
-                std::cout << (inst_res ? "SUCCESS" : "FAILURE") << std::endl;
-                res = res && inst_res;
-                out_device_buffer_->SetZero();
-            }
-            else
-            {
-                std::cout << "Given conv problem is not supported by instance: \n\t>>>>"
-                          << op_ptr->GetTypeString() << std::endl;
-            }
-        }
-        return res;
-    }
-
-    template <typename OpInstancePtr>
-    ProfileBestConfig Profile(const std::vector<OpInstancePtr>& op_ptrs,
-                              bool time_kernel     = false,
-                              bool do_verification = false,
-                              bool do_log          = false)
-    {
-        ProfileBestConfig best_config;
-
-        for(auto& op_ptr : op_ptrs)
-        {
-            auto invoker  = op_instance_.MakeInvokerPointer(op_ptr.get());
-            auto argument = op_instance_.MakeArgumentPointer(
-                op_ptr.get(), in_device_buffers_, out_device_buffer_);
-            if(op_ptr->IsSupportedArgument(argument.get()))
-            {
-                std::string op_name = op_ptr->GetTypeString();
-                float avg_time = invoker->Run(argument.get(), StreamConfig{nullptr, time_kernel});
-
-                std::size_t flops     = op_instance_.GetFlops();
-                std::size_t num_btype = op_instance_.GetBtype();
-                float tflops          = static_cast<float>(flops) / 1.E9 / avg_time;
-                float gb_per_sec      = num_btype / 1.E6 / avg_time;
-
-                std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                          << " GB/s, " << op_name << std::endl;
-
-                if(avg_time < best_config.best_avg_time)
-                {
-                    best_config.best_op_name    = op_name;
-                    best_config.best_tflops     = tflops;
-                    best_config.best_gb_per_sec = gb_per_sec;
-                    best_config.best_avg_time   = avg_time;
-                }
-
-                if(do_verification)
-                {
-                    out_device_buffer_->FromDevice(out_tensor_->mData.data());
-                    if(!ref_output_)
-                    {
-                        throw std::runtime_error(
-                            "OpInstanceRunEngine::Profile: Reference value not availabe."
-                            " You have to provide reference function.");
-                    }
-                    // TODO: enable flexible use of custom check_error functions
-                    CheckErr(out_tensor_->mData, ref_output_->mData);
-
-                    if(do_log) {}
-                }
-                out_device_buffer_->SetZero();
-            }
-        }
-        return best_config;
-    }
-
-    void SetAtol(double a) { atol_ = a; }
-    void SetRtol(double r) { rtol_ = r; }
-
-    private:
-    template <typename F, std::size_t... Is>
-    void CallRefOpUnpackArgs(const F& f, std::index_sequence<Is...>) const
-    {
-        f(*std::get<Is>(in_tensors_)..., *ref_output_);
-    }
-
-    template <std::size_t... Is>
-    void AllocateDeviceInputTensors(std::index_sequence<Is...>)
-    {
-        (AllocateDeviceInputTensorsImpl<Is>(), ...);
-    }
-
-    template <std::size_t Index>
-    void AllocateDeviceInputTensorsImpl()
-    {
-        const auto& ts = std::get<Index>(in_tensors_);
-        in_device_buffers_
-            .emplace_back(
-                std::make_unique<DeviceMem>(sizeof(std::tuple_element_t<Index, InArgsTypesTuple>) *
-                                            ts->mDesc.GetElementSpaceSize()))
-            ->ToDevice(ts->mData.data());
-    }
-
-    static constexpr std::size_t kNInArgs_ = std::tuple_size_v<InTensorsTuple>;
-    const OpInstanceT& op_instance_;
-    double rtol_{1e-5};
-    double atol_{1e-8};
-
-    InTensorsTuple in_tensors_;
-    TensorPtr<OutDataType> out_tensor_;
-    TensorPtr<OutDataType> ref_output_;
-
-    DeviceBuffers in_device_buffers_;
-    DeviceMemPtr out_device_buffer_;
-
-    template <typename T>
-    bool CheckErr(const std::vector<T>& dev_out, const std::vector<T>& ref_out) const
-    {
-        return ck::utils::check_err(dev_out, ref_out, "Error: incorrect results!", rtol_, atol_);
-    }
-};
-
-} // namespace utils
-} // namespace ck
diff --git a/library/include/ck/library/utility/ranges.hpp b/library/include/ck/library/utility/ranges.hpp
index 55c322f1ace041b9c4f346a5df4250c2323759dd..f11e4204a0cec5378328f69d3bb553a6227c47a3 100644
--- a/library/include/ck/library/utility/ranges.hpp
+++ b/library/include/ck/library/utility/ranges.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
index cc8787458d7ffff072cf8c055c8def43ca487e51..e730e9f589faf8047dd60a218ef191eea510cb63 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
index 04200cfb52e26f227d89852e2845547e6624c22d..f6696ffa9f6ca2c0665ecde3f0394c0802588a46 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
index 7b86f3cc728fdac1117cbe927445b6830e81e9ab..32d6f258b5c9152205f8b13e0326739b567b0845 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
index 2afb1afbc3d2355f682af9cad21b62c0ab1ee8ea..ee246ba56edd7271d7d2f980c62003a7ab7c7f06 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
index 68d768949c859f5ca8aedfebfa888253dc6c276c..5a9483b3099d9985f88f0121135a79d55edefd5a 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
index 737e5bfca3bccbafcd8179d80bb43b58bd545639..0fa0719237b000b7ab9b13d94cfa951edacf3139 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
index e09d01736083f54cf2c1b848814b2d7fb683d3b8..2f42f62b0b14b91bf94208b27684ba4b0b10c5d2 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
index 984d66e2884273e006772530e6a978891766a8e0..10b4cea7d7cddd4e694c0de42d122bb7ff3bd505 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
index 12cada9c44d61d8b81ad2dfa4877dc57b26ed681..c687eb20f06918b41ca54fa72a2a577b809eb2cf 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
index 13f198862e5583b1f63c51efaab11fa733976cfe..b19374ca6d5635238da838808d863b19fd1b6b93 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
index 2ca1adc2f6d160a97a2f40b4075f07b185e68351..bbd318ba9299005910d63a83d6906f4b69571479 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
index fe5de52796f6c9a0c94a84d883a7b1a714c4a652..187ccb5ff62fab57a83139f340a4032fd03e8dea 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
index 5b55c8e15e01d879cc11e57128c3e2dfdf81890f..ec2b2646ffa1a78b058a90b0508eba7f1d65a2be 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp
index 9517e4577e5b9af188aaf07413e8c0b3cf6c32ee..d76cd350c81807f330f7b04ae2ed2d1bf6bb5080 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
index 43b91244060ed90c9f2558dc99279ce0a086845b..ef65106c2cf537ae29dcc7313829ecaf3ae0ed58 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
index 326500fcbf013e8540fff9252202ac7af33b2938..078b241f9dfe194778806824911d791c7dd7a2de 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index e1bfa88f49d4fb886f953661c03b8638516ba646..4db05589b153b2307d377e67147b5ee1bf8336bc 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
index f59b742534b2c234b2b4908f5e540f591b104619..e25f903a8272ceb6bd559b8edfdb72f7ac759716 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..cd9c95c066e6f14c4ee6904255e333db1792dba9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
@@ -0,0 +1,4 @@
+add_instance_library(device_batched_gemm_bias_permute_instance
+    device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
+)
+
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a0afaabbc7bbff785c7a3e6cba884a0bb1f768e4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
@@ -0,0 +1,98 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16       = ck::half_t;
+using F32       = float;
+using F16_Tuple = ck::Tuple<F16>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+static constexpr auto ABSpec         = ck::tensor_operation::device::TensorSpecialization::Packed;
+static constexpr auto DESpec         = ck::tensor_operation::device::TensorSpecialization::Default;
+
+// A[g0, m0, m1, k0] * B[g0, n0, n1, n2, k0] + D[g0, m0, m1, n0, n1, n2] = E[g0, n0, m0, n0, n1, m1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_batched_contraction_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_mnnm_instance =
+    std::tuple<
+        // clang-format off
+        //############################################| NumDimG| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM|              A|              B|             DE| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################################|        |        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Spacialization| Spacialization| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################################|        |        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |               |               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################################|        |        |        |        |      |      |        |         |          |      |             |            |            |               |               |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        //M1 faster dim
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceBatchedContractionMultipleD_Xdl_CShuffle<       1,       2,       3,       1,   F16,   F16,     F32,      F16, F16_Tuple,   F16,  PassThrough, PassThrough,         Add, GemmMNKPadding,         ABSpec,         ABSpec,         DESpec,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        // clang-format on
+        >;
+
+void add_device_batched_contraction_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_mnnm_instance(
+    std::vector<std::unique_ptr<DeviceBatchedContractionMultipleD<1,
+                                                                  2,
+                                                                  3,
+                                                                  1,
+                                                                  F16,
+                                                                  F16,
+                                                                  F16_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  Add>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_batched_contraction_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_mnnm_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 9b96194c874c3daea7c9c4eb931368ce5645b50e..67dfc4cd35066053c7b9936a76c402454cf1e99b 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
index 0713dfcd99ca183d8f64cd23ac6067be990b39e7..9001c901cfbecb473c6a0e4f70a9db6526e2a4cb 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fda55a930363b43a91235d39c3b37cb028fdc380
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
@@ -0,0 +1,18 @@
+add_instance_library(device_batched_gemm_multi_d_instance
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b20e8a86b1a2474a08a1186f65dfe93adafb44fd
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e98fd24eeb759fb8bdde4d6982cdc219522e065d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,84 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..75042280f04c16982516c45d92ad46d40d98e91a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c8e1acfd4f8a4a00f4e727657a5a7df327c57527
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f1ed7a69423a1abfb92b85368a921472db5fc5cf
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..6e25517ed8600b8df740b14fa67b9ab9c60bdec3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..5a9fd16c83b9b7e44e718da96eb7612f9ac039be
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..cf5502926b772e670ff8a2136eeafc17e548da8e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..43cdcc5f140fa792181893b4bc2e586c9350e087
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..617877499d1c97bf99535f13ff057946487032af
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 4, 4>,      S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,      S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1f58aecf5084b0993fc6f8e75e7b8a406cad9382
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e0ead0ec4940482bbd19a68affd9064d09d9cd83
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..55fa86f37ff58b82a881cce16750cb78c19e828d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b8af3e61a3ae6ce06f50ed475d4f40162839e851
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..29d20e58ddbd3ad646d107dec278a4821936b35d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..dd0f8749c00dceeabc834787dfb7c907c3ede909
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
index 521c3d9219b1cd23e30ca194d3e22d1f38fa9316..cb89d3cefd6525d017d7f7b4685be456cb1c7e44 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
index 231d612d781bc93c2a232f7e1478fddfb09c1b5d..91eefba0c10af87e7674cb053ee549a2799de7a9 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
index 165bc3957d36479b1cb3052c0085ade58fb71322..c20798f557e58e870f748d7900f1418f3a1a5779 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
index 832fc3b066066f9ac8faae30cebff714b73b53c9..3d9ad64b9b40a2271ed40608735bffb0133153f9 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 99e8712474996a6c30e8bf16ae69f5bc15925364..cf23d01bf234e2cdda3efb2b655879a4740f5c27 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index 76121ffc347e0b8019c2ed1d18b6e71a5b0fb518..eba248e59978c867d53d64a6e43e9f2187a2e7d5 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -1,5 +1,7 @@
 add_instance_library(device_batched_gemm_softmax_gemm_permute_instance
     device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
     device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+    device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+    device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
 )
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..498bf58fb3e6cfaf444fc92923ff261389ac0ac8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
@@ -0,0 +1,135 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F32  = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAdd    = ck::tensor_operation::element_wise::ScaleAdd;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmPadded  = ck::tensor_operation::device::GemmSpecialization::MNKOPadding;
+
+static constexpr auto TensorDefault = ck::tensor_operation::device::TensorSpecialization::Default;
+
+// c[g, m, n] = a[g, m, k] * b[g, n, k]
+template <index_t NumDimG,
+          index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          index_t NumDimO,
+          MaskingSpecialization MaskingSpec>
+using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances =
+    std::tuple<
+        // clang-format off
+        // #############################################|  NumDimG| NumDimM| NumDimN| NumDimK| NumDimO|  AData|  B0Data|  B1Data|  CData|     Acc0BiasData| Acc1BiasData| AccData| CShuffle|           A|          B0|        Acc0|          B1|           C|           GEMM|   ATensorSpec|  B0TensorSpec|  B1TensorSpec|   CTensorSpec| NumGemmK| Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| MaskingSpec|  D0s Bias|
+        // #############################################|         |        |        |        |        |   Type|    Type|    Type|   Type|             Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            | SrcScalar|
+        // #############################################|         |        |        |        |        |       |        |        |       |                 |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            | PerVector|
+        // #############################################|         |        |        |        |        |       |        |        |       |                 |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |          |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,        1>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,    64,    32,   8,   8,    2,   32,   32,     1,     8,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,   128,    32,   8,   8,    2,   32,   32,     1,     8,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    32,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    32,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    32,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    32,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        // Padded fallback kernel
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,        1>,  
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
+        // clang-format on
+        >;
+
+void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
+    std::vector<std::unique_ptr<
+        DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                            1,
+                                            1,
+                                            1,
+                                            1,
+                                            BF16,
+                                            BF16,
+                                            BF16,
+                                            BF16,
+                                            ck::Tuple<BF16>,
+                                            ck::Tuple<>,
+                                            PassThrough,
+                                            PassThrough,
+                                            ScaleAdd,
+                                            PassThrough,
+                                            PassThrough,
+                                            MaskingSpecialization::MaskOutUpperTriangle>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances<
+            2,
+            1,
+            1,
+            1,
+            1,
+            MaskingSpecialization::MaskOutUpperTriangle>{});
+}
+
+void add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
+    std::vector<
+        std::unique_ptr<DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            BF16,
+                                                            BF16,
+                                                            BF16,
+                                                            BF16,
+                                                            ck::Tuple<BF16>,
+                                                            ck::Tuple<>,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            ScaleAdd,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            MaskingSpecialization::MaskDisabled>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances<
+            2,
+            1,
+            1,
+            1,
+            1,
+            MaskingSpecialization::MaskDisabled>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..744bd6456d28f719a28508f65bdbd78bb02d4120
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -0,0 +1,137 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAdd    = ck::tensor_operation::element_wise::ScaleAdd;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmPadded  = ck::tensor_operation::device::GemmSpecialization::MNKOPadding;
+
+static constexpr auto TensorDefault = ck::tensor_operation::device::TensorSpecialization::Default;
+
+// c[g, m, n] = a[g, m, k] * b[g, n, k]
+template <index_t NumDimG,
+          index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          index_t NumDimO,
+          MaskingSpecialization MaskingSpec>
+using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances =
+    std::tuple<
+        // clang-format off
+        // #############################################|  NumDimG| NumDimM| NumDimN| NumDimK| NumDimO| AData| B0Data| B1Data| CData|    Acc0BiasData| Acc1BiasData| AccData| CShuffle|           A|          B0|        Acc0|          B1|           C|           GEMM|   ATensorSpec|  B0TensorSpec|  B1TensorSpec|   CTensorSpec| NumGemmK| Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| MaskingSpec|  D0s Bias|
+        // #############################################|         |        |        |        |        |  Type|   Type|   Type|  Type|            Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            | SrcScalar|
+        // #############################################|         |        |        |        |        |      |       |       |      |                |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            | PerVector|
+        // #############################################|         |        |        |        |        |      |       |       |      |                |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |          |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,       1>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+ #if CK_WORKAROUND_SWDEV_388832
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,    64,    32,   8,   8,    2,   32,   32,     1,     8,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+ #endif
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,   128,    32,   8,   8,    2,   32,   32,     1,     8,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    32,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    32,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    32,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    32,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        // Padded fallback kernel
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,       1>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
+        // clang-format on
+        >;
+
+void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
+    std::vector<std::unique_ptr<
+        DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                            1,
+                                            1,
+                                            1,
+                                            1,
+                                            F16,
+                                            F16,
+                                            F16,
+                                            F16,
+                                            ck::Tuple<F16>,
+                                            ck::Tuple<>,
+                                            PassThrough,
+                                            PassThrough,
+                                            ScaleAdd,
+                                            PassThrough,
+                                            PassThrough,
+                                            MaskingSpecialization::MaskOutUpperTriangle>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances<
+            2,
+            1,
+            1,
+            1,
+            1,
+            MaskingSpecialization::MaskOutUpperTriangle>{});
+}
+
+void add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
+    std::vector<
+        std::unique_ptr<DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            1,
+                                                            F16,
+                                                            F16,
+                                                            F16,
+                                                            F16,
+                                                            ck::Tuple<F16>,
+                                                            ck::Tuple<>,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            ScaleAdd,
+                                                            PassThrough,
+                                                            PassThrough,
+                                                            MaskingSpecialization::MaskDisabled>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances<
+            2,
+            1,
+            1,
+            1,
+            1,
+            MaskingSpecialization::MaskDisabled>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
index 53ad7ba5ffa9272da26189a45a75a9453af68f48..b342612d1c098070108696b7a76a8e7fe754fb51 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -45,6 +45,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_
         // #############################################|         |        |        |        |        |   Type|    Type|    Type|   Type|         Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
         // #############################################|         |        |        |        |        |       |        |        |       |             |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
         // #############################################|         |        |        |        |        |       |        |        |       |             |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,    64,    32,   8,   8,    2,   32,   32,     1,     8,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
@@ -58,8 +59,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         // Padded fallback kernel
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
         // clang-format on
         >;
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 21da6895e66be8597fdeda60501e2d25716f3221..3fd0c07370d40407dd3ef754f55ccff9deb091df 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -45,9 +45,12 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_
         // #############################################|         |        |        |        |        |  Type|   Type|   Type|  Type|         Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
         // #############################################|         |        |        |        |        |      |       |       |      |             |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
         // #############################################|         |        |        |        |        |      |       |       |      |             |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+ #if CK_WORKAROUND_SWDEV_388832
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,    64,    32,   8,   8,    2,   32,   32,     1,     8,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+ #endif
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,   128,    32,   8,   8,    2,   32,   32,     1,     8,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    32,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
@@ -58,8 +61,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         // Padded fallback kernel
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
         // clang-format on
         >;
 
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batchnorm/CMakeLists.txt
index d12a2f244fc8b0dedc42ac716ea8354a5c1dee72..19a3cc8cd1c59babc813efb73cbe26b3ec09db3a 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/CMakeLists.txt
@@ -7,4 +7,8 @@ add_instance_library(device_batchnorm_instance
     device_batchnorm_backward_f32_instance.cpp
     device_batchnorm_backward_bf16_instance.cpp
     device_batchnorm_backward_f64_instance.cpp
+    device_batchnorm_infer_f16_instance.cpp
+    device_batchnorm_infer_f32_instance.cpp
+    device_batchnorm_infer_bf16_instance.cpp
+    device_batchnorm_infer_f64_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp
index b62c8b99cbdb6cf42922e843c11a0832614b6ea7..bc95d2f1b1f1ad3734daaa19ed5dbb1faff5a6a4 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp
index d05b8b592c29aceef387a4fdec8c02854501995f..fbc8d0bc6097a477fe9920201167a58142aa9b44 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp
index e3ef95d12e17acfebc029ac6e1a878194691739f..bed38658a99c109370e6271dc94f785d94268106 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp
index 41be396c24a28b5a7fbc47b10d7cf6ae5e6e07e9..fc5ec77e428dfbfb09a902245f27ca25b8bc0d35 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp
index cd1e05b1133ad2a2515a936ef968667a03bc11be..4e38ee13b24dd150f1366ed7694e69c6969a2e2c 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp
index 073dd583f97e0eed167e4bc4ef952530e82854a2..f087eb79824e3e36c6bc07a3d42b68f38a89c16b 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp
index be63bd44c66c3204c66990ff6048bd895841c84f..d0f361401a500c16f1228d4df1d3de6714d1f43a 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp
index fe87091e8d66e32af0832c1b013abce0911673b2..710d07b8280cef467d2fac8a70034129feb13fda 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..8801c309f9138ccf321ba1dcac5f1718287e0a76
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp
@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F32  = float;
+
+using Normalize = ck::tensor_operation::element_wise::NormalizeInInfer;
+
+// clang-format off
+template <index_t Rank>
+using device_batchnorm_infer_bf16_instances =
+     std::tuple <
+        // Tuple<XDataType, MeanDataType, VarDataType, ScaleDataType, BiasDataType>, Tuple<YDataType>, NormalizeOp, Rank, MPerThread, Sequence<XVectorSize, MeanDataType, VarDataType, ScaleVectorSize, BiasVectorSize>, Sequence<YVectorSize> 
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 1, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 2, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 2, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 2, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 2, Sequence<2, 2, 2, 2, 2>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<2, 2, 2, 2, 2>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<4, 1, 1, 1, 1>, Sequence<4> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<1, 4, 4, 4, 4>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<4, 2, 2, 2, 2>, Sequence<4> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<2, 4, 4, 4, 4>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, Rank, 4, Sequence<4, 4, 4, 4, 4>, Sequence<4> >
+     >;
+// clang-format on
+
+void add_device_batchnorm_infer_rank_4_bf16_instances(
+    std::vector<std::unique_ptr<
+        DeviceElementwise<Tuple<BF16, F32, F32, BF16, BF16>, Tuple<BF16>, Normalize, 4>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_batchnorm_infer_bf16_instances<4>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b674cfc423d81b5efce31169e1c81741ba156b9a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp
@@ -0,0 +1,54 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Normalize = ck::tensor_operation::element_wise::NormalizeInInfer;
+
+// clang-format off
+template <index_t Rank>
+using device_batchnorm_infer_f16_instances =
+     std::tuple <
+        // Tuple<XDataType, MeanDataType, VarDataType, ScaleDataType, BiasDataType>, Tuple<YDataType>, NormalizeOp, Rank, MPerThread, Sequence<XVectorSize, MeanDataType, VarDataType, ScaleVectorSize, BiasVectorSize>, Sequence<YVectorSize> 
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 1, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 2, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 2, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 2, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 2, Sequence<2, 2, 2, 2, 2>, Sequence<2> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<2, 2, 2, 2, 2>, Sequence<2> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<4, 1, 1, 1, 1>, Sequence<4> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<1, 4, 4, 4, 4>, Sequence<1> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<4, 2, 2, 2, 2>, Sequence<4> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<2, 4, 4, 4, 4>, Sequence<2> >,  
+        DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, Rank, 4, Sequence<4, 4, 4, 4, 4>, Sequence<4> >  
+     >;
+// clang-format on
+
+void add_device_batchnorm_infer_rank_4_f16_instances(
+    std::vector<std::unique_ptr<
+        DeviceElementwise<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, 4>>>& instances)
+{
+    add_device_operation_instances(instances, device_batchnorm_infer_f16_instances<4>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..05e3650887c3dfa84d1f03a8380d1144c7886acf
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp
@@ -0,0 +1,52 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/utility/tuple.hpp"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F32 = float;
+
+using Normalize = ck::tensor_operation::element_wise::NormalizeInInfer;
+
+// clang-format off
+template <index_t Rank>
+using device_batchnorm_infer_f32_instances =
+     std::tuple <
+        // Tuple<XDataType, MeanDataType, VarDataType, ScaleDataType, BiasDataType>, Tuple<YDataType>, NormalizeOp, Rank, MPerThread, Sequence<XVectorSize, MeanDataType, VarDataType, ScaleVectorSize, BiasVectorSize>, Sequence<YVectorSize> 
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 1, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 2, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 2, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 2, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 2, Sequence<2, 2, 2, 2, 2>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<2, 2, 2, 2, 2>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<4, 1, 1, 1, 1>, Sequence<4> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<1, 4, 4, 4, 4>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<4, 2, 2, 2, 2>, Sequence<4> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<2, 4, 4, 4, 4>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, Rank, 4, Sequence<4, 4, 4, 4, 4>, Sequence<4> >
+     >;
+// clang-format on
+
+void add_device_batchnorm_infer_rank_4_f32_instances(
+    std::vector<std::unique_ptr<
+        DeviceElementwise<Tuple<F32, F32, F32, F32, F32>, Tuple<F32>, Normalize, 4>>>& instances)
+{
+    add_device_operation_instances(instances, device_batchnorm_infer_f32_instances<4>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..15a02af021a22b4e9e43186a7b3e18b938534dd8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/utility/tuple.hpp"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64 = double;
+
+using Normalize = ck::tensor_operation::element_wise::NormalizeInInfer;
+
+// clang-format off
+template <index_t Rank>
+using device_batchnorm_infer_f64_instances =
+     std::tuple <
+        // Tuple<XDataType, MeanDataType, VarDataType, ScaleDataType, BiasDataType>, Tuple<YDataType>, NormalizeOp, Rank, MPerThread, Sequence<XVectorSize, MeanDataType, VarDataType, ScaleVectorSize, BiasVectorSize>, Sequence<YVectorSize> 
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 1, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 2, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 2, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 2, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 2, Sequence<2, 2, 2, 2, 2>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 4, Sequence<1, 1, 1, 1, 1>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 4, Sequence<2, 1, 1, 1, 1>, Sequence<2> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 4, Sequence<1, 2, 2, 2, 2>, Sequence<1> >,
+        DeviceElementwiseImpl<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, Rank, 4, Sequence<2, 2, 2, 2, 2>, Sequence<2> >
+     >;
+// clang-format on
+
+void add_device_batchnorm_infer_rank_4_f64_instances(
+    std::vector<std::unique_ptr<
+        DeviceElementwise<Tuple<F64, F64, F64, F64, F64>, Tuple<F64>, Normalize, 4>>>& instances)
+{
+    add_device_operation_instances(instances, device_batchnorm_infer_f64_instances<4>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
index ffd6a6a7be2081429508fbbaec0edcf6bb1dbf39..d2a0a3d0fbe7a869b2d1603fe8879a201c1ca4b5 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
@@ -1,7 +1,13 @@
 add_instance_library(device_contraction_bilinear_instance
+    #float
     device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
     device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
     device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
     device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
+    #double
+    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
+    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
 )
 
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
index ebbff88346b9dcb1eb4716b1b36f456b806ef1ca..5587db77e08ceb40242084853ed07236cc992605 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
index 980383f3e71b3f1058c295c9aba55c56f596134a..26262855ea8b6e3ba2c929e527e97dc5108495dd 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
index 2d4b6e3489a9ccce619b0a76dc097f04d72b0d57..befc0dcd10d07735ec046e8ef4d64bb54be4756c 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
index 7caa469f54b0169449250f000df038820e35a32f..e45b47cf94ed53de17b862e50ed2ef7a00e3e978 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f437a227d50b665b999e30b2d065ac6825e7ec04
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
@@ -0,0 +1,76 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..13fdbeb35c53a93947f739dd26f91885cff4a72b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
@@ -0,0 +1,76 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..95ef8c4929b00263c65034ac1ec48195efdc8fff
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
@@ -0,0 +1,76 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..290f81d7c9a926dd4e9c169e3515ebdbaa547788
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
@@ -0,0 +1,76 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
index 7ad6605486caa7fb4e60e717db6756f52f453bd8..31f6a0fcdc9a04d15d63ffc93e2dac550a0ed293 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
@@ -1,7 +1,13 @@
 add_instance_library(device_contraction_scale_instance
+    #float
     device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
     device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
     device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
     device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
+    #double
+    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
+    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
+    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
+    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
 )
 
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
index 5118d0d033ee733c9b3bd7bc3524824c77aae667..16fd1cb407bf81741845c41baa580f208e98508d 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
index 655d4f0061ae1c417e610e9059f7e2bbd2048363..ff37bf7cceb40f75c8a54309341914539057d678 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
index a9d20be18bf5c52044b12f84c58f7824f8ef8142..8a1f6f93341edbb585157cd66308ab8f781bfedc 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
index a68f5c9718a1e547c62f16888a4fd0a0ed1d8197..d333f597268d55a195da671fcbcdaab2c03acfc0 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..4c87b51a9e426945f049d471bf3591d1f222b72b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..fd3f57c6b6058f16e9e34158fefc71e3fb9c5f3c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1e53f0b2f51f23143b1d49d1b7d9b2d3e2b7b765
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d02d146a9bf98b25fe7de56148c8879dd62a328b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
index 5a5c8384227150dd8717433dce6c9ea2395896bd..e3e90c966dd6efd920ea0147d434f926bd07bd05 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
index e0f3d6199f8939bb0f1ce7f8efc48f2a92cfd3c6..81e9122d9506eb3ae78602c7beedbd2ce1b0b745 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
index 30537d9373b8175207b267666574033cae4e1a3c..dbc82168f49ad102886ce8aebaea72cd3ef5a99a 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
index 190c39b870b3cb33e9080c8b591884f4ec387347..3ac250f3e6a9d3fe8b3139aef5a79dc520269381 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
index e14cd558628772991da9b9d4ebb3dd5d852e3285..fcb858728e80c2cce14579f5a35409316deceeb4 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
index f001b83c1711a7185d71f5b2c7f795005b35b696..2baa6ae061473a44e962c597d2fabb9d62cabee4 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
index 83ba6a1c6bb8a3098da1b31a59a1d4a12b8b062e..28867fe19dd62f63d2e3df798a210a1045415fb8 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index 1da9a81d904b3e43aaf5ace4f612c43c05e66ee0..40656e382b8074580de72fed78123c07b5e1e303 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
index 7c33df5e7687ee3e279b350bec9ca17902e84231..422e37e926fd7a839afc2cf884b141ca32737078 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
index a5f8629f2daef05ca1dbcbec2e8c4c668bf96a84..5993f6bd7a28c2992bf2e069880e84bf75612dfd 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index 8076d6d35669114d84b922a59948b888dcdab506..bb9b696864c54265a8d3d119bcd55d82452a825e 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
index 33503b9f8ae00a92c25b178629736c7e8a5474c3..da96c79a6e9a44edd05928b54a865483f6a1e1a2 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index c5e4bd199ea8322312af832b63a0ea559109131a..78e9c893c9e82a723d1794685c7d33d78129047d 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
index f43d13e309333bc86abf22465b7f035c1445321b..4663a9ac32ac3b6424d638c2d6bc7248651c240a 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
index 0ce6b04c4218190ccd3cad238f1bc90ec3002df9..0b1df52f8c1ce4503c154c4f7ce40b713455e839 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index 76ab3189d7f210446b305a3daec80a2902de63ff..9969a8bbc90df83a31da50f8f064cbe0a83923f7 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp
index f8c255088877e95e2cbbbacbfba6f9e1f4fc0291..e34ea06ff46ba090a212575e66dcfd454a44de66 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp
index fe7152471ed54aa5334e820f993232912013c6de..3254fcfc26467edc3ec6b14d1838e784d7438529 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
index 04ce7c07639c968d90a8053f4b67b1335acc9b69..94b2a47e50b635e5b862c109f479660183d2bede 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
index 0251d9157f3b894208b17091ab6c79eec8bf605d..4244ab7b8770c9cbd60957c4fd66f004b7f5975c 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
index c2975727e48385207678abdcbc1fda776e761664..5c7db4ca3bc4dce89cd1d139722d1e91758f8d7a 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
index fc86d7302450a01263dc07d8f895c669795f3644..ebc56487a159afe8b32e5e321c6996c38250c32f 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp b/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
index baddecf6451dbc58f5ee46a2d5fc5fcf754d609e..f2a5f0728ac1d8bc4bff78aa5476e872a7085163 100644
--- a/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
@@ -1,12 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -28,17 +28,25 @@ using Normalize = ck::tensor_operation::element_wise::Normalize;
 using device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances = std::tuple<
     // clang-format off
     //###################|<in, mean, square_mean, gamma, beta>| <out>|  functor| NDim| MPerThread| <in, mean, square_mean, gamma, beta ScalarPerVector>| <out ScalarPerVector>|
-    DeviceElementwise<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   8,       Sequence<8, 1, 1, 8, 8>,      Sequence<8>                >,
-    DeviceElementwise<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   4,       Sequence<4, 1, 1, 4, 4>,      Sequence<4>                >,
-    DeviceElementwise<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   2,       Sequence<2, 1, 1, 2, 2>,      Sequence<2>                >,
-    DeviceElementwise<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   1,       Sequence<1, 1, 1, 1, 1>,      Sequence<1>                >
+    DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   8,       Sequence<8, 1, 1, 8, 8>,      Sequence<8>                >,
+    DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   4,       Sequence<4, 1, 1, 4, 4>,      Sequence<4>                >,
+    DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   2,       Sequence<2, 1, 1, 2, 2>,      Sequence<2>                >
+    // clang-format on
+    >;
+
+using device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_generic_instance = std::tuple<
+    // clang-format off
+    DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   1,       Sequence<1, 1, 1, 1, 1>,      Sequence<1>                >
     // clang-format on
     >;
 
 void add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(
-    std::vector<DeviceElementwiseBasePtr<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, 2>>&
+    std::vector<DeviceElementwisePtr<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, 2>>&
         instances)
 {
+    add_device_operation_instances(
+        instances, device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_generic_instance{});
+
     add_device_operation_instances(
         instances, device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances{});
 }
diff --git a/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp
index 7f15372ed91b7adbb4cbec369613f6c4c500d59e..3e2386ee03558ec6d346abc73a037e65e2d9cde3 100644
--- a/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp"
@@ -23,6 +23,11 @@ template <typename XElementwise, typename YElementwise, index_t Rank, index_t Re
 using device_elementwise_normalization_f16_instances =
     std::tuple <
         // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
+        DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel for large N
+        DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel for large N
+        DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>, // fallback kernel for large N
+        DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 2, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>, // fallback kernel for large N
+        DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 4, 64, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel for large N
         DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel
         DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel
         DeviceElementwiseNormalizationImpl<ck::Tuple<F16, F16>, F16, F16, F32, F16, XElementwise ,YElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 4, 1, 4, 1, 4, 4>, // fallback kernel
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index e20d592c84ed8da549f29160c87f65579307096a..11227d01f5b2247c1330fb296c5aa768897c3612 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -1,43 +1,115 @@
 add_instance_library(device_gemm_instance
-   device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+   device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
+   device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
+   device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
+   device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
+   device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
    device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
    device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
    device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
    device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
    device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
    device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
+
+   device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
+   device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
+   device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
+   device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
+)
+
+set(ENABLE_PIPELINE_V2_OPT OFF)
+
+if (ENABLE_PIPELINE_V2_OPT)
+set(MAX_ILP_OPTS
+   -mllvm
+   -amdgpu-enable-max-ilp-scheduling-strategy
+)
+set(WAVES_PER_EU_DEFS
+   CK_USE_WAVES_PER_EU=1
+   CK_MIN_WAVES_PER_EU=1
+   CK_MAX_WAVES_PER_EU=1
+)
+set(IGLP_OPT_DEFS
+   CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT=1
 )
+
+# layout=NT
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS ";;"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+# layout=NN
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+# layout=TT
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS}")
+# layout=TN
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+endif(ENABLE_PIPELINE_V2_OPT)
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
index 5d2f18e14e37be6b1c5528e8f43f23d1b6c88996..ea99a5a30ef80bdb954df78535a6cd63b8663bbb 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_km_kn_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d444e29aa33538874b306caeab0fd78e177ce807
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
index 01e3b3793a487fd5eb6b6b5a6ca6a10e86bd0588..b83acfa8cb4db572bb33ccd20e6d38e7594f0234 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_km_nk_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f8f8a0bd333748899bc52eed1f65ff6c56467a87
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
index 804e86a066ffa47c6d96de81cd4bef5644a5b131..d5800e03336e9161e01ef240a46aefbf9b13c2c0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..840a4fabe1c211689796debc86e8ce63e0c19b4e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
index 159fa90f7432248721f3bfef606f6df3b04584e9..abe52ce1da0edb191755d205a4e744c274ae56b1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -35,7 +35,39 @@ using device_gemm_dl_f16_f16_f16_mk_nk_mn_instances =
         //  ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         //  ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         //  ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
         // clang-format on
         >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..82b1b5dc29fc4e5612236f2e0271597e405dfe80
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances =
+    std::tuple<
+        // clang-format off
+        //  ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        //  ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        //  ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        //  ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        // clang-format on
+        >;
+
+void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
index d8e7798438dcaaa9d7c2e48e8074a2f58a635438..e696bfdcdc07f6799a00b9463837031559b715a5 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
index 0034ac59c3888d8dcc1b2bf8e136ac4f02c2e6ad..d3ad7c60ec6f8cf2321aaa03e18e62ac28e65773 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
index 0b540b8b3492c355d75adf98cacc1826417db0b9..a56a36b0ab9cf0260a0366560d8816f18f1f5d7e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
index 4f6ff5111b1fa5dbf39e4c60d1770a319aa9dbba..63d55e81d6a5cb1157ea72a97e01599e6fc750ff 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
index a4208245e5756ccf3a2e6be29c3a5c5f73e97acc..3d9a265c29b35f09699fb12b578942a5c5f9ed89 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_km_kn_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0439201511f285a0d7eba405df72d1942a4fb49b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 4, 4>,      S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,      S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
index 06fab7f68a2efef857107e3ef6038b4505e21712..240384d19c5eeebb86ac92be53d22605b1862801 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_km_nk_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..350834f7e515b93645e2f22285c19f9b9285304c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
index b6d72fa221f4b946f3a25363a99b737a8a830428..e96905247d6a72e057bf13d821fdc17f78a380cf 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_mk_kn_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..27397527bc973c56bf15a8c35e86dec89184959c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
index 67d2e3ce4b41c097fffb3099b97a8d398b44b743..124b818b28cfe1145a5936a9ea77500b980df0ad 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_mk_nk_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b99f3f2b63902c4b832c86dcdc3f74376d972dbe
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
index 03eebf4ec3062cf5f85486288673fb96fb2761b1..2e884dfc8afba8d6c6ea88a8e3159c7a62660364 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
index 5d8de04cd9fa66931dde4ed0494fc429659506c3..2ca29b1e6f0791154d16ec1b5dd134d8d9c77c37 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
index 7b12b7cf1f42da6d8c321c1ac121166753b7ff59..706076098da912ea9e5a844ccb2797cbc850a01b 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
index 730ffd4633de3c72cb6ab932126a18d8798418dd..5ac458a7b985d16faf90a53866ec4f9d2bc9c6b0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
index 619473ff0ce40a5150c59efdfd62bc905d7ce83f..a64412544721237542c729d0f9581acb39917555 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
index 7d5ecb5dc046e35d3f96e93af70b597dbb286df1..5895f1f3f279b7a1f6632fe02e02b99b9b70d8e5 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
index f9458b7483c5c29011b4aa063061436da23f9e36..23176269c2c84808ad05a874259694f22c6064d1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
index 77a03b746cc12e5f8cea8772c4432c9fb4e24110..31a9abe53a2312c25ebeb5157e81998054d2345b 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
index ef8d7d4e40ef4de46857db4932e109780e52f044..201fd93110b604b1c6df849fcc5c88ff98a93bea 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
index cb65cc7b68b359ec015af3adcd3a7866983d492b..5d489b207efdf62de64d99469acfdbe49a13859f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
index 5b1014ed827d08f275b5da4c20abe14f153b9950..e09480d57bc20e54697fcd3c5ae761a66fda678b 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
index e6f6add8bf45e98f6e9607b48a924e00ef642323..34065c334d1bee6064718a8dbac6f0b6f51c5826 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
index 80b3d03da26c49136c8220f8a433d8798ad751e1..95d7777a79371c4d9dfa1ec6ff09b16dc55a3807 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
index 93b3df1e572604e70d059c6cb622f8143eb2a92a..bf24bc76b1b8e3c64473df1d6b985712c53f4a20 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
index f10365d892455e3e36f5e05ec94b7a4f577c1fbc..023f987127cc61983fb11f53838cff9383f0f257 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
index a7a9eb62cef20c05cee9ff76e5d1c0b21fb95ab9..ffb199e58fe3a0200197ef1d119feec348a01502 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
index 9fb45b00365896be88e6aff214922c51325a4d65..90e979d891a5c4d1d9ca18653311afeda9029e5b 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/common.hpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..de96a22630c88ccfe9a4b1a2f84b01e0d4751613
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/common.hpp
@@ -0,0 +1,43 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/data_type.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+using InstanceNT = DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+using InstanceNN = DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+using InstanceTT = DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+using InstanceTN = DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+
+template <typename Instance>
+using OwnerList = std::vector<std::unique_ptr<Instance>>;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d887b16e6015ae2bdf0b4a157e6c5f8bef75b5d7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceNT;
+using Instances = OwnerList<InstanceNT>;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..81fedd50f0b2c790fe93de9891dc0f952d9cb798
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+        // clang-format off
+        // pipeline v1, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..5a0c52c2df9efde9d755420ffa34bf21bd1522f1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..59ffb80bd4b6fcbbd7850ec017c526f41a87f87c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|           LoopScheduler|             Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                        |                     |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                        |                     |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                        |                     |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a64424e8acb300fe51d2c44c0ebd02a92e731c6d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e313a790e10b0f0c94c491e0409d630cb6441938
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1, LoopScheduler::Default,          PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..050ff6e231eede6fc5b83fa4a779007f44634115
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1, LoopScheduler::Default,          PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0a0406baec7c7fe883ff1877132fff16a07ad3cd
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|            LoopScheduler|                   Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                         |                           |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                         |                           |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                         |                           |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1, LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..7fb1a4845db11028c35639cd0e78fbb387e195ea
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceNN;
+using Instances = OwnerList<Instance>;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a0dd60c0f56a7d985d9e65e5419753aefba775c6
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+        // clang-format off
+        // pipeline v1, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..122fff4960306d94839161f4c12b0828f95446ea
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..7fc55b017e8247247fbcc66bd143a3b94bf54681
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|              Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                      |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..3671bea7a3fc8fd8240976d616bb63b34fd7f642
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d7e7a0550515134f5a0f5f739ad731183cb6aa4b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..15e191e5a81c329eb98d698a9dd118f54459d61f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..95fc8ecb46be3107cf47566526c67e2ddf4a35c2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e3a7ccefb4d73fadad67f0e93b58f93ff3ead54f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceTT;
+using Instances = OwnerList<Instance>;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..98db8bad1ca01f32c78786e778ba302002259b1f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+        // clang-format off
+        // pipeline v1, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..532c348b7e1571bb7d64b33ac29ab4d270dc1f22
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..5314c7fc20371a0924f578752761c9663cd1ade0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|              Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                      |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..fa53a3bf0f689376c9752c0cb32915408acb71d8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..bc4b3e801bc52a64122f0cf14998b7a6ae2c047b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0c8e8b2c416f724166184889b794283155f1ee59
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c9d1913aec737f581990403b5b4ec5768c718978
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..70bfd0a80cb7f7ce9335396e2fbbdf6613f6b67d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceTN;
+using Instances = OwnerList<Instance>;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0f4ccceb50a187daf14ec85a0ba5e59e877b0538
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..134755299d58073e28a1320ca26a08a2dd981c3f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1aba1fa8d99192b8a59ad33dd3ca49261b0a8105
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|           LoopScheduler|             Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                        |                     |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                        |                     |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                        |                     |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0410eabb700eeed4ae642d2a8f788df2b0ea9467
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e6e73cdccaf9095aac0b5e88915416df41a9f1bb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..78f18fd5821450a5cf701feb7ebaaca0d443579d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a41919aab7c84d9a2436445869baf47c3ec5c531
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
deleted file mode 100644
index 18a78674e7ac4d79e6da3b035c8e9ea966618f16..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
+++ /dev/null
@@ -1,110 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_km_kn_mn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
-        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-        // clang-format on
-        >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_km_kn_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
deleted file mode 100644
index cef6070af8a72a6cdb0616574736e722094f3a74..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+++ /dev/null
@@ -1,110 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_km_nk_mn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-        // clang-format on
-        >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_km_nk_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
deleted file mode 100644
index 1be70d6ca414f785ec2582fd90bc5d3d2617e4e3..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ /dev/null
@@ -1,137 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-        // clang-format on
-        >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
deleted file mode 100644
index 6b8455ffa93fcceaf9d0e5d5595197f8bafddcbe..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ /dev/null
@@ -1,130 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
index b9e28e3d7ef70b2e3aa785b048280fb4959ce312..a590413accba4e9e0dddd8df2952cd9d3fdb8b35 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
index 2b1a5a57bb8b22b9235fdd02fe3f9da659b13c0d..1d010d1b0792dfc62f0a1de235558ab8e4e71fd0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
index 301d3b55b59c0a6818b27570fe1c9a7b9ff44946..f108b753421598c376eefb8cfa52b0a4f1522920 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
index cd16f35ff233fb04c79d50888f54b806e79e611a..b0b4bc012df347cf3e487d96ecef2e03826f5e77 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
index 39166698473dbd05010f960e82366012ffaa1863..df3bd94fcafa5062b597584c3aadf2772c376f39 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
index 0a623034ef65a1cfd0d220b9ad2253ab0b298620..73b4e776668c969fb6ddcd38f74bd1a74fea48c8 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
index 5ef8d08de90a65ae04f552fc87447e3b7887bc92..76137a1c3e1e2cb82c9e15ec9bc71e4abe76391e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
index c9557bae893e1affc0f2138aaa9cfca206e267d1..f0158d8f3d51fbb773988d02c14c329e9243ebf9 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
index 463e0865c0ae6980c534005d1ace62750645a74c..125fbc21a2b9c141adc8d11b0727667ca062fada 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[k, m], b[k, n], d0[m, n], d1[m, n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -139,6 +150,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
index b71ff1b9972a97276a0fedf0106814a54ba19a0c..cc33692d7378178903fe2b0a9a43180571d70276 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[k, m], b[n, k], d0[m, n], d1[m, n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -139,6 +150,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
index 9060c9b1b084512f038ccd5c5b0c3573ac32f0a9..704787a080ee0af7ea74f9f75a290a69cf73adf0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[k, n], d0[m, n], d1[m, n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -139,6 +150,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
index 81cf01d6a9d70346e8ea2387220fb03258e6e580..d64c9ec5e15aa85c54d5c852755be63ca0597327 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[n, k], d0[m, n], d1[m ,n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -130,6 +141,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
index 4da85cc46eb4ded15bcd63dd7b0d89864637d2b0..e68bd8e7e4aa032dc56007e22aba0f90ca0c6af8 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0)
 // outout: e[m, n]
 // input: a[k, m], b[k, n], d0[m, n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -123,6 +134,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
index ab83e4baabf80242bfbf569fb13b5d39a3686ad7..5aaa2e8fe509a18ac360859f0cdcf67535b75fd8 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[k, m], b[n, k], d0[m, n], d1[m, n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -123,6 +134,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
index a4cd3fadbe93f4231388e9efccbe28a2c69a6f1b..7a2a3dbaf3d82cbed71e91e2bebe8fb3f3dd4ea0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[k, n], d0[m, n], d1[m, n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -123,6 +134,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
index 207e76ffe5f18f9a4d64a0083c27829933e7a5fa..fa3360997886de27f75b89457d8e1ba9ae4471d1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[n, k], d0[m, n], d1[m ,n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -114,6 +125,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..eb9345cbadb28905a73cde8035e245c914a1bd15
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/CMakeLists.txt
@@ -0,0 +1,6 @@
+add_instance_library(device_gemm_add_multiply_instance
+   device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
+   device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
+   device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
+   device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d9ae7817914dbcb7dcae38254afef6f02e54a4da
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
@@ -0,0 +1,106 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16       = ck::half_t;
+using F32       = float;
+using F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddMultiply = ck::tensor_operation::element_wise::AddMultiply;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // no padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|           CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|   Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|     Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |              |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   256,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   256,    32,   2,   2,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    64,    32,   2,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    64,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,    64,    32,   2,   2,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,    64,   128,    32,   2,   2,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        // M/N/K Padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|           CDE|            GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|   Elementwise|  Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|     Operation|                |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |              |                |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   2,   2,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   2,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   2,   2,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   2,   2,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+        // clang-format on
+        >;
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..27c6cbe8ab71398f9c3bc03bb977c71af397836d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
@@ -0,0 +1,106 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16       = ck::half_t;
+using F32       = float;
+using F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddMultiply = ck::tensor_operation::element_wise::AddMultiply;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // no padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   256,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   256,    32,   2,   8,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    64,    32,   2,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    64,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,    64,    32,   2,   8,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,    64,   128,    32,   2,   8,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        // M/N/K Padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   2,   8,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   2,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   2,   8,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   2,   8,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+        // clang-format on
+        >;
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ff5f0e94a1732e3b36bc9295c4f1ce9a24fa467f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -0,0 +1,106 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16       = ck::half_t;
+using F32       = float;
+using F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddMultiply = ck::tensor_operation::element_wise::AddMultiply;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // no padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        // M/N/K padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+        // clang-format on
+        >;
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..dba625e0ada1d4f09cb26a5860849a707a35cb97
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -0,0 +1,143 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16       = ck::half_t;
+using F32       = float;
+using F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddMultiply = ck::tensor_operation::element_wise::AddMultiply;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // no padding
+        // N % 8 == 0 && K % 8 == 0
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+
+        // M/N/K padding
+        // N % 8 == 0 && K % 8 == 0
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+
+        // M/N/K padding
+        // N % 4 == 0 && K % 4 == 0
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+
+        // M/N/K padding
+        // N % 8 == 0 && K % 1 == 0
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    AddMultiply, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>
+
+        // clang-format on
+        >;
+
+void add_device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    AddMultiply>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..97693a2566fc440eb8a7e71bd02251700191dfda
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
@@ -0,0 +1,6 @@
+add_instance_library(device_gemm_add_relu_add_layernorm_instance
+   device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
+   device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
+   device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
+   device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..28a452c1a1124a326337cf4c21183035294fda1d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
@@ -0,0 +1,130 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16           = ck::half_t;
+using F32           = float;
+using F16_F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row           = ck::tensor_layout::gemm::RowMajor;
+using Col           = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddReluAdd  = ck::tensor_operation::element_wise::AddReluAdd;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e = elementwise((a * b), d0, d1)
+// h = layernorm(e, gamma, beta)
+// outout: h[m, n]
+// input: a[k, m], b[k, n], d0[m, n], d1[m, n], gamma[n], beta[n]
+template <LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances = std::tuple<
+    // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|            CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|          PostShuffle|     PostShuffle|            Layernorm|       Layernorm|     LoopScheduler|     Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise|    Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|      Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                 _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |               |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                     |                |                     |                |                  |             |
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   2,   2,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   2,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   2,   2,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   2,   2,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+// irregular tile size
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|            CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|           PostShuffle|     PostShuffle|            Layernorm|       Layernorm|          LoopScheduler|              Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise|    Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|  ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                       |                      |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|      Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                  _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                       |                      |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |               |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                      |                |                     |                |                       |                      |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,  PipelineVersion::v1>
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+        // pipeline v1, 2 waves
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Interwave, PipelineVersion::v1>
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+        // pipeline v2, 1 wave
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,     AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Col,
+                                                             Row,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances<
+            LoopScheduler::Interwave,
+            PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v2>{});
+#endif
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..13366238d6d9008d7ed425485670df5229f71cee
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
@@ -0,0 +1,130 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16           = ck::half_t;
+using F32           = float;
+using F16_F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row           = ck::tensor_layout::gemm::RowMajor;
+using Col           = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddReluAdd  = ck::tensor_operation::element_wise::AddReluAdd;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e = elementwise((a * b), d0, d1)
+// h = layernorm(e, gamma, beta)
+// outout: h[m, n]
+// input: a[k, m], b[k, n], d0[m, n], d1[m, n], gamma[n], beta[n]
+template <LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances = std::tuple<
+    // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|         CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|          PostShuffle|     PostShuffle|            Layernorm|       Layernorm|     LoopScheduler|     Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                 _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                     |                |                     |                |                  |             |
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   2,   8,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   2,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   2,   8,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   2,   8,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+// irregular tile size
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|         CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|          PostShuffle|     PostShuffle|            Layernorm|       Layernorm|          LoopScheduler|               Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                       |                       |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                 _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                       |                       |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                     |                |                     |                |                       |                       |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,              S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,   PipelineVersion::v1>
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+        // pipeline v1, 2 waves
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,              S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Interwave, PipelineVersion::v1>
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+        // pipeline v2, 1 wave
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,              S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,   PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Col,
+                                                             Col,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances<
+            LoopScheduler::Interwave,
+            PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v2>{});
+#endif
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..8a4889ee83eabcae6466bdd5dd67d268c9d54287
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -0,0 +1,130 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16           = ck::half_t;
+using F32           = float;
+using F16_F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row           = ck::tensor_layout::gemm::RowMajor;
+using Col           = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddReluAdd  = ck::tensor_operation::element_wise::AddReluAdd;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e = elementwise((a * b), d0, d1)
+// h = layernorm(e, gamma, beta)
+// outout: h[m, n]
+// input: a[k, m], b[k, n], d0[m, n], d1[m, n], gamma[n], beta[n]
+template <LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances = std::tuple<
+    // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|         CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|          PostShuffle|     PostShuffle|            Layernorm|       Layernorm|     LoopScheduler|     Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                 _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                     |                |                     |                |                  |             |
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+// irregular tile size
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|         CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|          PostShuffle|     PostShuffle|            Layernorm|       Layernorm|          LoopScheduler|              Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                       |                      |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                 _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                       |                      |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                     |                |                     |                |                       |                      |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,  PipelineVersion::v1>
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+        // pipeline v1, 2 waves
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Interwave, PipelineVersion::v1>
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+        // pipeline v2, 1 wave
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Row,
+                                                             Row,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances<
+            LoopScheduler::Interwave,
+            PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v2>{});
+#endif
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..fc3cbcf9055892bfcf63dcf6d3d3c4515026ca71
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -0,0 +1,127 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16           = ck::half_t;
+using F32           = float;
+using F16_F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row           = ck::tensor_layout::gemm::RowMajor;
+using Col           = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddReluAdd  = ck::tensor_operation::element_wise::AddReluAdd;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e = elementwise((a * b), d0, d1)
+// h = layernorm(e, gamma, beta)
+// outout: h[m, n]
+// input: a[k, m], b[k, n], d0[m, n], d1[m, n], gamma[n], beta[n]
+template <LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances = std::tuple<
+    // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|         CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|          PostShuffle|     PostShuffle|            Layernorm|       Layernorm|     LoopScheduler|     Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                 _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                  |             |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                     |                |                     |                |                  |             |
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 4>,               8,             S<16, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 8>,               8,             S<32, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<32, 4>,               8,             S<32, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 8>,               8,             S<16, 8>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 4>,               8,             S<16, 4>,               1, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,             S<16, 4>,               8,             S<16, 4>,               1, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+// irregular tile size
+using device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //#######################################|      A|      B|            Ds|      H| AData| BData| AccData| CShuffle|        DsData| EMeanVarData| GammaData|  BetaData| HData|           A|           B|         CDE|           H|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|          PostShuffle|     PostShuffle|            Layernorm|       Layernorm|          LoopScheduler|              Pipeline|
+        //#######################################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|         Type|      Type|      Type|  Type| Elementwise| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| ThreadClusterLengths| ScalarPerVector| ThreadClusterLengths| ThreadSliceSize|                       |                      |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |   Operation|   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|                 _M_N|   _NWaveNPerXdl|                 _M_N|              _M|                       |                      |
+        //#######################################|       |       |              |       |      |      |        |         |              |             |          |          |      |            |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                     |                |                     |                |                       |                      |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,  PipelineVersion::v1>
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+        // pipeline v1, 2 waves
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Interwave, PipelineVersion::v1>
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+        // pipeline v2, 1 wave
+        ,
+        DeviceGemmMultipleDLayernorm_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,          F16,       F16,       F16,   F16, PassThrough, PassThrough,  AddReluAdd, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<16, 4>,              1,             S<16, 4>,               1, LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleDLayernorm<Row,
+                                                             Col,
+                                                             Row_Row_Tuple,
+                                                             Row,
+                                                             F16,
+                                                             F16,
+                                                             F16_F16_Tuple,
+                                                             F16,
+                                                             F16,
+                                                             F16,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             AddReluAdd,
+                                                             PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances<
+            LoopScheduler::Interwave,
+            PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instances<
+            LoopScheduler::Default,
+            PipelineVersion::v2>{});
+#endif
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
index e8747af4828a2492f6db8e23b575de4efb7ce90e..bfb95bce85cb8696e645ec6b5626abf84456916f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
index ed54c3a9bf58346a8829f3d7f275c6668e00ed1d..d0352339cf38fecf9c3b3d4fd8a91729ed53495f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
index da7eae637bfb16580b4ec98e479768c2649adad7..d5b298ab217b4269beca858f0d54e319bf6dbf4f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
index 34345095e0c2489f44437ea58f58b2891cc91ff1..80c8f018f416154d0b2ccd26cb9f8822233a43e4 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
index 55461dfba783de45a2991836535b3b652cc49a44..74ec9e1f8d28a77b0f78181cd6c05ea93b140e16 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
index 405e69975ca9ffec97999bce38fd73d953e139ed..eb98b3e7e871a1c5f4129032031175443ddc4de0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
index 9af31b3a129d2d1bf1ff6f32d0ee8e252cbdfa29..5f4a90125ac827ccfa406b5bf81f11313ce54258 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
index 841b7a1d47ce48b8486f0af04ff80571e4601d1a..38e3897d6a552d8a2f797b9f3b4baaa98636a53d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
index 9f7f643beb6832594bdc393777e5e08663a1f94e..803c44c7f53352109c28510c75f59e98165d6d8e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[k, m], b[k, n]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -122,6 +132,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
index c8e9f35d240da45fb902a2d02cd73d2eab5167e1..9b9ef3db24cb28e0783b5003678beeab01dd22f1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[k, m], b[k, n]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -122,6 +132,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
index 5f804d45a560ac3b7b519cb3ab1840214e392d39..1a0b6c9d1c62d1aeb378ca140fd09959d213533a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[m, k], b[k, n]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -122,6 +132,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
index 60cb138f5654bb6f2b10ea5ba0a03191a49906ab..18b1c0e993131b62d0692769090facd14034ea8e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[m, k], b[n, k]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -113,6 +123,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
index 59e2b2da8641b9fb094750077d760e9f611d9c98..f0e7b6ab43c51abeb4e79e9e30b104bf05f82a07 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
index bb09bf8b8e82f6c2056881f37b9e65c73657f0e5..56815b9ac5fa4d89b649089c73fbfeddc985cc8e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
index 0a3b566de64505adb19f75ffe5e126cfed5da438..e66d46a26995ec521df8fe15a7fd01277ebaff0d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
index 2b17e47b1cce153654d74ca7497b5c04117071de..fb1dfac69cd525ee7b0caa3d3d7b732ba5042bd8 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
index e178d3b0adeb563c36d6a8f1f3936d048df23f75..fed2cbbfb980ab27e8810b643874fc9b308b21b7 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
index 52be9fe709bd710f3e05f381e342deef5c7affae..44ac4c08cd97881f98f0c36a9bd5d42cd5078327 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index 9b5ff404845d5ae00b5664ee44a405e979ce52d2..30a2bf36d1cef771764e2832fc8d8a91b7120588 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -26,7 +26,8 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+// static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
@@ -35,14 +36,22 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
index 7fc35c4198eb49bf2625f6cfa8eb80a001fdc086..4b2a2dbdc73aee32a427f91470610325a5bb3080 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
index f27b2199e0c9dcb1e91566cb0aa0a05edaef5b0e..9d15ccd362774c77a9613046995d44761b8ccb49 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
index b9a1095570aaf33aa8f3f93710cbb6d3ae276430..4e9ad58742c552f437dd4b8174cb675984dfc681 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
index 44e5f597d0d714438c97a9ce7d591736f4fe7f48..330e5aff907af00920dcc4f3cb1f5f2a3b50e6fa 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
index f3a9063f7fb847b9c303ea836f97ed4cf3173ab9..0db3a15d2a4816cca200820346ccf7c9a711fa31 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
index 05ba449246e8c6f68182eb707a3e9da0974e2078..25ea4f48cf8cd82bda19c28411d5745bc78713d7 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -1,85 +1,21 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = bhalf_t;
-using F32  = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
-using device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -92,12 +28,22 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_insta
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_bf16_f32_bf16_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_bf16_f32_bf16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<1,
+                                                                    GNWC,
+                                                                    GKXC,
+                                                                    GNWK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       1,
+                                       GNWC,
+                                       GKXC,
+                                       GNWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
index 7a610a747cc9cc7555bd18b17fd499d19cd7c439..8065012f1258179601dc9befd9c11a778f6e2cf6 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -1,84 +1,21 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
-using device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f16_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-using device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f16_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -91,11 +28,22 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f16_default_instances{});
-    add_device_operation_instances(
-        instances, device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<1,
+                                                                   GNWC,
+                                                                   GKXC,
+                                                                   GNWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       1,
+                                       GNWC,
+                                       GKXC,
+                                       GNWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
index 90e074f052c6f935bcbf862250670289e0634617..c70a54c2dc5664ff863e83f6072361e69c9f86b2 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -1,83 +1,21 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
-using device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f32_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f32_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-    // clang-format on
-    >;
-
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -90,11 +28,22 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f32_default_instances{});
-    add_device_operation_instances(
-        instances, device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f32_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<1,
+                                                                   GNWC,
+                                                                   GKXC,
+                                                                   GNWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       1,
+                                       GNWC,
+                                       GKXC,
+                                       GNWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
index 74aebf1031d5b02db54fece32fa693e0c532287d..5aa50adb31ecf5c8c636d0bb34946b166c5919e6 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
index 361ea8f4ee9ff8144fd2392754fdaa53f380e40b..333b40c71ce1c755f7d7862c9ddae8013392ea8a 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
index 3145b716402e4e5cf8e542dc991bc7db0a655e74..506a93ae9ba07d7a53e770809f2c47399432e476 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
index cde93f902c95d1c28f85b7a7627089fe3e97f841..30084f16d1d4c86ee2778b0d3526c2e593d39259 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
index 3b2968d48efd3e0061c36df1febb290b64264f73..85ec0f55aa78db969e8731213094f786a281d8d1 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
@@ -1,3 +1,8 @@
 add_instance_library(device_grouped_conv2d_bwd_data_instance
    device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..01cd23b2e3b1b857e1f2f629ba92f0a49efea3b9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<GNHWK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          GNHWC,
+                                                          ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<GNHWK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          GNHWC,
+                                                          ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 3d604d42cc3a9fcbf8e9750ab5e3b2064f0bc161..bb0fd36e44f16987b3a11f7d8699a8510566fd59 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,80 +1,14 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
-
-static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
-
-using device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances = std::tuple<
-    // clang-format off
-        // 1. Default
-        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-
-        // 2. Filter1x1Stride1Pad0
-        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise|              ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|               DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                                 |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                                 |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
-    // clang-format on
-    >;
-
+// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -89,8 +23,22 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances)
 {
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
     add_device_operation_instances(
-        instances, device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances{});
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b671017f51c8dc31b256a95f1214ed17fb03fc63
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..bfe6e47d19218519f243ade0e36d8ea0c8567be8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
@@ -0,0 +1,141 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
+
+static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
+    ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+// f16_f16_f32_f16
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv2d_bwd_data_xdl_f16_instances = std::tuple<
+    // clang-format off
+        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
+        
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#endif
+    // clang-format on
+    >;
+
+// bf16_bf16_f32_bf16
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv2d_bwd_data_xdl_bf16_instances = std::tuple<
+    // clang-format off
+        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   BF16,   BF16,     F32,      BF16, Empty_Tuple,   BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
+        
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
+#endif
+    // clang-format on
+    >;
+
+// f32_f32_f32_f32
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv2d_bwd_data_xdl_f32_instances = std::tuple<
+    // clang-format off
+        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  F32,  F32,      F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>
+
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#endif
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0b1fa63ddb3ce300db54be0beeade361f86bd2cc
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<NHWGK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          NHWGC,
+                                                          ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<NHWGK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          NHWGC,
+                                                          ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1c8fc57e1c778be8b25426b2a62f06767b609b45
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..fcdfe8d42bf05aaeee6909380f3413ac9ae9b16a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
index 4009121e7fb12084e94308e3a1b1070ce9f72b11..b7b9fc92d1263f4537ebdfe976b811d6c45757cb 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
@@ -2,5 +2,8 @@ add_instance_library(device_grouped_conv2d_bwd_weight_instance
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
 )
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index ede21f1f4f7124156540518bf4252d3526929606..cf39c8601033c41e80f6ed6340d76d874ee5a6cd 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -1,85 +1,16 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = bhalf_t;
-using F32  = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
-using device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -92,12 +23,22 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_in
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_bf16_f32_bf16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<2,
+                                                                    GNHWC,
+                                                                    GKYXC,
+                                                                    GNHWK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 99e556618c3bfc535807ae8595a33e0f6b6d5390..d52f0b4d1c6bb99d63852b62b3c54cef29f3221b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,84 +1,16 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
-using device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f16_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-using device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f16_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -91,12 +23,22 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f16_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<2,
+                                                                   GNHWC,
+                                                                   GKYXC,
+                                                                   GNHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 15871a28c3a2a85c053fc3259f9de67b5694ede6..62547a5014d527f87dc242fe41cd2e157bd93617 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,83 +1,16 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
-using device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f32_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f32_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-    // clang-format on
-    >;
-
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -90,12 +23,22 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f32_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f32_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<2,
+                                                                   GNHWC,
+                                                                   GKYXC,
+                                                                   GNHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1cb9991a86cbeecc12fe8813630596299d8f35cb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<2,
+                                                                    NHWGC,
+                                                                    GKYXC,
+                                                                    NHWGK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e64d55c3bbb986cbf3bc314d1a4bd4e28b5a663e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<2,
+                                                                   NHWGC,
+                                                                   GKYXC,
+                                                                   NHWGK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..5fa4c9ba3cd1482a81c5aa0ed942737cc3e60b2a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<2,
+                                                                   NHWGC,
+                                                                   GKYXC,
+                                                                   NHWGK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
index 5ef1b6866efce4ade692d86e50a3240fdd6ac0ae..a36e1b47caa6c35082d64523941134ecc9ffb320 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
@@ -3,11 +3,11 @@ add_instance_library(device_grouped_conv2d_fwd_instance
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
    # NHWGC, GKYXC, NHWGK
+   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
    device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
-   #dl 
+   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+   #dl
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..85a7a5be257365f37234c3b723d47a86a253b24d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
@@ -0,0 +1,53 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index fc18b3c7358c44b3bdbf00db99274f432beefea4..5eb881549cf37c7aba2dae0e5ac1aa3402a9c3a9 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,100 +1,54 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "device_grouped_conv2d_fwd_dl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using InDataType  = ck::half_t;
-using WeiDataType = ck::half_t;
-using AccDataType = float;
-using OutDataType = ck::half_t;
-
-using Empty_Tuple = ck::Tuple<>;
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-using InLayout  = ck::tensor_layout::convolution::GNHWC;
-using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-using OutLayout = ck::tensor_layout::convolution::GNHWK;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-static constexpr auto Filter1x1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-static constexpr auto Filter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Pad0_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,  Filter1x1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Stride1Pad0_instances =
-    std::tuple<
-        // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|          Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|              Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation|       Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |                     |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp, Filter1x1Stride1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              InLayout,
-                                                              WeiLayout,
+                                                              GNHWC,
+                                                              GKYXC,
                                                               Empty_Tuple,
-                                                              OutLayout,
-                                                              InDataType,
-                                                              WeiDataType,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
                                                               Empty_Tuple,
-                                                              OutDataType,
-                                                              InElementOp,
-                                                              WeiElementOp,
-                                                              OutElementOp>>>& instances)
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances{});
+                                   device_grouped_conv2d_fwd_dl_f16_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
 
-    add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f16_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
 
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Stride1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f16_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 648b39637ebe7a49d42c5dd2620834b76a04cc9b..4157853c41767f8ef705c864482fa56724d16e19 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,104 +1,54 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "device_grouped_conv2d_fwd_dl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using InDataType  = float;
-using WeiDataType = float;
-using AccDataType = float;
-using OutDataType = float;
-
-using Empty_Tuple = ck::Tuple<>;
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-using InLayout  = ck::tensor_layout::convolution::GNHWC;
-using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-using OutLayout = ck::tensor_layout::convolution::GNHWK;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-static constexpr auto Filter1x1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-static constexpr auto Filter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances = std::tuple<
-    // clang-format off
-           // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Pad0_instances = std::tuple<
-    // clang-format off
-           // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,  Filter1x1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Stride1Pad0_instances =
-    std::tuple<
-        // clang-format off
-           // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|          Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|              Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation|       Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |                     |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp, Filter1x1Stride1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              InLayout,
-                                                              WeiLayout,
+                                                              GNHWC,
+                                                              GKYXC,
                                                               Empty_Tuple,
-                                                              OutLayout,
-                                                              InDataType,
-                                                              WeiDataType,
+                                                              GNHWK,
+                                                              F32,
+                                                              F32,
                                                               Empty_Tuple,
-                                                              OutDataType,
-                                                              InElementOp,
-                                                              WeiElementOp,
-                                                              OutElementOp>>>& instances)
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances{});
+                                   device_grouped_conv2d_fwd_dl_f32_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
 
-    add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f32_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
 
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Stride1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f32_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
deleted file mode 100644
index 1cb5d06999512687cc1ef0e9fee1444096bcf1c2..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
+++ /dev/null
@@ -1,104 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using InDataType  = int8_t;
-using WeiDataType = int8_t;
-using AccDataType = int32_t;
-using OutDataType = int8_t;
-
-using Empty_Tuple = ck::Tuple<>;
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-using InLayout  = ck::tensor_layout::convolution::GNHWC;
-using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-using OutLayout = ck::tensor_layout::convolution::GNHWK;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-static constexpr auto Filter1x1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-static constexpr auto Filter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Pad0_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,  Filter1x1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Stride1Pad0_instances =
-    std::tuple<
-        // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|          Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|              Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation|       Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |                     |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp, Filter1x1Stride1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-        // clang-format on
-        >;
-
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              InLayout,
-                                                              WeiLayout,
-                                                              Empty_Tuple,
-                                                              OutLayout,
-                                                              InDataType,
-                                                              WeiDataType,
-                                                              Empty_Tuple,
-                                                              OutDataType,
-                                                              InElementOp,
-                                                              WeiElementOp,
-                                                              OutElementOp>>>& instances)
-{
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances{});
-
-    add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Pad0_instances{});
-
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Stride1Pad0_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..631c2ce0939f7ac0948563a1c8a83dae8f0eb2db
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "device_grouped_conv2d_fwd_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
+    // clang-format off
+           // ########################################|        NDim| InData| WeiData|    MultpleD| OutData| AccData| InLayout| WeiLayout| MultipleD| OutLayout|          In|          Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+           // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+           // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+           // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
+    // clang-format on
+    >;
+
+template <typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_dl_f32_instances = std::tuple<
+    // clang-format off
+        // clang-format off
+           // ########################################|        NDim| InData| WeiData|    MultpleD| OutData| AccData| InLayout| WeiLayout| MultipleD| OutLayout|          In|          Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+           // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+           // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+           // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index 29f3310314548b332d6fab41f7ddcebb215775eb..bc7d577b69b623901b71e4127ee67d55676947f2 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -1,137 +1,14 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using BF16 = ck::bhalf_t;
-using F32  = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // OddC
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,    64,    32,   8,   8,   32,   32,    4,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    1,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -146,8 +23,41 @@ void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 6a4a3d2a45e2a7c11cedab653bf4bed447537115..55bf6da9b9e1b876bce918ee8d611c69a053d820 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,137 +1,14 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // OddC
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,    64,    32,   8,   8,   32,   32,    4,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    1,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -147,7 +24,40 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances{});
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 1fec35fd9fffa59afe9c7bdc81e3dd7ada2dde8f..202cdd6b446b20277eccedcd89bbf2346b906241 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,109 +1,14 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -119,7 +24,40 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances{});
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..07bea1c03cafca9e7cd3069c1fb04c0e1d6cf9f4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
@@ -0,0 +1,105 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "device_grouped_conv2d_fwd_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_xdl_f16_instances =
+    std::tuple<
+        // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        // clang-format on
+        >;
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_xdl_bf16_instances =
+    std::tuple<
+        // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|            GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |                |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |                |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        // clang-format on
+        >;
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_xdl_f32_instances =
+    std::tuple<
+        // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|            GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |                |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |                |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
+        // clang-format on
+        >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..82edf896db1c932c2d3b19d9361e1294801b4035
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              BF16,
+                                                              BF16,
+                                                              Empty_Tuple,
+                                                              BF16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
index 8aca730433060d045e9c2045e6a366b08e0fe5cf..4bd3236caf31392c6626b0df299d02082f55c6e1 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -1,137 +1,14 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NHWGC = ck::tensor_layout::convolution::NHWGC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using NHWGK = ck::tensor_layout::convolution::NHWGK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
-using device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // OddC
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,    64,    32,   8,   8,   32,   32,    4,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    1,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -147,7 +24,40 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances{});
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..4f5bdb202349ad0a5c10bbde97caefda0eca61f2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
index e48db4a5314528bb45f4444f93da4ba8872e86d8..c8f456db836ef3cc43198188edb30b0676bf0046 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -1,85 +1,21 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = bhalf_t;
-using F32  = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
-using device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -92,12 +28,22 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_bf16_f32_bf16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<3,
+                                                                    GNDHWC,
+                                                                    GKZYXC,
+                                                                    GNDHWK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       3,
+                                       GNDHWC,
+                                       GKZYXC,
+                                       GNDHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
index 1655850ec148568734cbdbace8eefe8cb6487e0c..099123ecbc7695c2cdd3bd1c8e28113aab4d0227 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -1,85 +1,21 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
-using device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f16_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-using device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -92,12 +28,22 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f16_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<3,
+                                                                   GNDHWC,
+                                                                   GKZYXC,
+                                                                   GNDHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       3,
+                                       GNDHWC,
+                                       GKZYXC,
+                                       GNDHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
index aba46b7ebeb90afb8881fc7ad4d8a2e1281eb146..0eda980b30bae334904b2ab93935ca9e6703a04e 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -1,84 +1,21 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
-using device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f32_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f32_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -91,12 +28,22 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f32_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f32_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<3,
+                                                                   GNDHWC,
+                                                                   GKZYXC,
+                                                                   GNDHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       3,
+                                       GNDHWC,
+                                       GKZYXC,
+                                       GNDHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
index 78eedca5f7687d7adad6caa0dbb91da84550cd94..cd209dbf9e11e0031848fac0148f8e7e848ed480 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
@@ -3,4 +3,9 @@ add_instance_library(device_grouped_conv3d_fwd_instance
    device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
    device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
    device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
+
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
index b4ae8b6ce5f2ba3d8f3569449934d84fc631252c..c3b100ea69c8312dd72a724d8b9c4e242efa86df 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
index 061674bd829840def99f4dcf317886844d398fea..ca488e9dc429e7cbb943979ce79f9f2dda157c48 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
index ed7e5476760983e8ca5a17475585ec5eb9e834f2..6087b1b187cd4af0d45402379dc391479ea936c5 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
index bf5fa306013e485e6dd57faa011bee8217001148..fd8c47deb38ab31641622317c28c9d7303c85392 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d2cacf4539de6ed0f0e6b85397244f913bd1d823
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
@@ -0,0 +1,129 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
+using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances =
+    std::tuple<
+        // clang-format off
+        // Default
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+
+        // Filter1x1Pad0
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+
+        // Filter1x1Stride1Pad0
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        // clang-format on
+        >;
+
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              BF16,
+                                                              BF16,
+                                                              Empty_Tuple,
+                                                              BF16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..6354f011129b2128e28dd8e64940d102c31e04b7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -0,0 +1,129 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
+using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances =
+    std::tuple<
+        // clang-format off
+        // Default
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+
+        // Filter1x1Pad0
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+
+        // Filter1x1Stride1Pad0
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        // clang-format on
+        >;
+
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..381ac5ef087a5202a05388f19ccc7245832fdcdb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
@@ -0,0 +1,128 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F32 = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
+using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances =
+    std::tuple<
+        // clang-format off
+        // Default
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+
+        // Filter1x1Pad0
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+
+        // Filter1x1Stride1Pad0
+        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
+        // clang-format on
+        >;
+
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
similarity index 82%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
index 59b012134058cbd64a3548ea3892cd477660d4d9..6bd53d869343a0fdae7bd890a60ab854153b5a58 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -22,9 +22,9 @@ using Empty_Tuple = ck::Tuple<>;
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
+using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
@@ -39,74 +39,74 @@ static constexpr auto ConvFwd1x1S1P0 =
 
 static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
-// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances = std::tuple<
+// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances = std::tuple<
     // clang-format off
         // Default
         //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
 
         // Filter1x1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
                                                                                                                                                                                                                 
         // Filter1x1Stride1Pad0                                                                                                                                                                                 
         //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
     // clang-format on
     >;
 
-void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NDHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -116,7 +116,7 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances{});
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
index 82beb2ace282742a17a14881b0520677b1324c74..b973b70aacc86a5e5baadd2e5b72f17704754fa4 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
@@ -3,4 +3,8 @@ add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
    device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
    device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
index 5f2097b07bd4d767eb8d5fd4d6780f445d2db99e..aa161e51c989b2128a9a1e1c5ad055d2678ae737 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -65,7 +65,15 @@ using device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_tile_instances = st
         //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
         DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
-        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              2,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              2,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   2,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   2,   2,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   2,   2,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
index 677bd1a2c643993c4617bc3e29619d2df5dc7665..c454deac1a3a045f51a853d1546a70500360cc15 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -65,7 +65,15 @@ using device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_tile_instances = st
         //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
         DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
-        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   2,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   2,   8,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   2,   8,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
index 95a1a87d76337be22ea4c4335eeb2398ba774f3c..c829e8863d2c1156c1754ee3ae73c47a5f84a07a 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -64,8 +64,16 @@ using device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances = st
         //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              2,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
index a103406d53b5321d66a00474611d7630936855e1..fb30e7a97317aa74d589d0987011b54dc67dc62b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
@@ -61,8 +61,19 @@ using device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_tile_instances = st
         //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
-        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,    
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    32,   8,   8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..8642562fa38f25a7e487d7b4194ee8f8f16beb91
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[m, k] * b[k, n] = e[m, n]
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // Currently AK1 must equal BK1 !
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..83b31b07cf45c068be8711838ff43ab2ad0101a6
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -0,0 +1,87 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough                    = ck::tensor_operation::element_wise::PassThrough;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // Currently AK1 must equal BK1 !
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..aa6365cd98c039c36007b875d425e3cef37ba782
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough                 = ck::tensor_operation::element_wise::PassThrough;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[m, k] * b[n, k] = e[m, n]
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f4460b360be563123fdf7e02c43b49008dfc43ea
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -0,0 +1,81 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough                    = ck::tensor_operation::element_wise::PassThrough;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,    
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    32,   8,   8,   32,   32,    1,    4,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..648f2146cbb84bfba719d175f3934c01e3da76fa
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
@@ -0,0 +1,6 @@
+add_instance_library(device_grouped_gemm_fastgelu_instance
+   device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
+   device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f4086b6eac4dd74903fbddb955d5bd7f41ee748a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -0,0 +1,104 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using FastGelu    = ck::tensor_operation::element_wise::FastGelu;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[k, m] * b[k, n] = e[m, n]
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   256,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   256,    32,   2,   2,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,   128,    32,   2,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    64,    32,   2,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    64,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,    64,    32,   2,   2,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,    64,   128,    32,   2,   2,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+    // clang-format on
+    >;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              2,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    64,    32,   2,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    64,   128,    32,   2,   2,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,    64,    32,   2,   2,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    64,   128,    32,   2,   2,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instances{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d68eb76144579b073f8119d431910696e6b4af9e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -0,0 +1,104 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using FastGelu    = ck::tensor_operation::element_wise::FastGelu;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[k, m] * b[n, k] = e[m, n]
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   256,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   256,    32,   2,   8,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,   128,    32,   2,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    64,    32,   2,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    64,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,    64,    32,   2,   8,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,    64,   128,    32,   2,   8,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    16,    64,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 4>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,              1>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    64,    32,   2,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    64,   128,    32,   2,   8,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,    64,    32,   2,   8,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    64,   128,    32,   2,   8,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Col,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Col,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instances{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..2dfb8caace5144fa29934af425b005c28e9d77c1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,103 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using FastGelu    = ck::tensor_operation::element_wise::FastGelu;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[m, k] * b[k, n] = e[m, n]
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instances{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..598a0b0e29c22effb655b09247bc84ef6c1dd268
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,104 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using FastGelu    = ck::tensor_operation::element_wise::FastGelu;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[m, k] * b[n, k] = e[m, n]
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,    
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,   128,    32,   256,    32,   8,   8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough,    FastGelu, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  FastGelu>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instances{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
index aa0cc1148051c8000b588cc72d0f15f38a412112..176fb2fbee77418b877b0b76d0a2086dcce2da92 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
@@ -1,4 +1,11 @@
 add_instance_library(device_normalization_instance
-    device_normalization_f16_instance.cpp
-    device_normalization_f32_instance.cpp
+    device_layernorm2d_f16_instance.cpp
+    device_layernorm2d_f32_instance.cpp
+    device_layernorm4d_f16_instance.cpp
+    device_layernorm4d_f32_instance.cpp
+    device_groupnorm_f16_instance.cpp
+    device_groupnorm_f32_instance.cpp
+    device_groupnorm_swish_f16_instance.cpp
+    device_groupnorm_swish_f32_instance.cpp
+    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e3820462cf85ff5392e6e43a3fa78208425242e1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Pass = ck::tensor_operation::element_wise::PassThrough;
+
+void add_device_normalization_rank_5_3_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 5, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Pass, 5, 3>{});
+    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 5, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d85817aad3194f7c225d7d45da601c26a1046196
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Pass = ck::tensor_operation::element_wise::PassThrough;
+
+void add_device_normalization_rank_5_3_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 5, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Pass, 5, 3>{});
+    add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 5, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a81f776c0f36008f9acf72c6328e8d6cd59ee4d6
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Swish = ck::tensor_operation::element_wise::Swish;
+
+void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F32, F16, Swish, 5, 3>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_normalization_f16_f32_f32_f16_generic_instance<Swish, 5, 3>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_f32_f32_f16_instances<Swish, 5, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f4bb8bda8145f5e642f7028f7d9271e611038809
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Swish = ck::tensor_operation::element_wise::Swish;
+
+void add_device_normalization_rank_5_3_swish_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Swish, 5, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Swish, 5, 3>{});
+    add_device_operation_instances(instances, device_normalization_f16_instances<Swish, 5, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..bbb9bd0fe8b83d3ed01489e5ebabc4708bba8a71
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Swish = ck::tensor_operation::element_wise::Swish;
+
+void add_device_normalization_rank_5_3_swish_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Swish, 5, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Swish, 5, 3>{});
+    add_device_operation_instances(instances, device_normalization_f32_instances<Swish, 5, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..3f7e4aff1a2987de40568fce9fed880838302a39
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Pass = ck::tensor_operation::element_wise::PassThrough;
+
+void add_device_normalization_rank_2_1_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 2, 1>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Pass, 2, 1>{});
+    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 2, 1>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1f0db3a03666c03522559c4322bfe172206d6804
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Pass = ck::tensor_operation::element_wise::PassThrough;
+
+void add_device_normalization_rank_2_1_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 2, 1>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Pass, 2, 1>{});
+    add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 2, 1>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..cb9d72e61423666be1fe0f2d2ab3e85d86832556
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Pass = ck::tensor_operation::element_wise::PassThrough;
+
+void add_device_normalization_rank_4_3_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 4, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Pass, 4, 3>{});
+    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 4, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ed555b840da42bf6693f2c93cdce5f8f7dceb246
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Pass = ck::tensor_operation::element_wise::PassThrough;
+
+void add_device_normalization_rank_4_3_f32_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 4, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Pass, 4, 3>{});
+    add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 4, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_normalization_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_normalization_f16_instance.cpp
deleted file mode 100644
index 8994d9dcb6e799509e521cbbd7d77f9ba5984fd2..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/normalization/device_normalization_f16_instance.cpp
+++ /dev/null
@@ -1,65 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
-#include "ck/utility/data_type.hpp"
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Pass = ck::tensor_operation::element_wise::PassThrough;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-// clang-format off
-using device_normalization_f16_instances =
-    std::tuple <
-        // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 4, 1, 4, 1, 4, 4>, // fallback kernel
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 4, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 2, 1, 2, 1, 2, 2>
-    >;
-// clang-format on
-
-void add_device_normalization_rank_2_1_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 2, 1>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 2, 1>{});
-}
-
-void add_device_normalization_rank_4_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 4, 3>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 4, 3>{});
-}
-
-void add_device_normalization_rank_5_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 5, 3>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 5, 3>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_normalization_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_normalization_f32_instance.cpp
deleted file mode 100644
index 4a7e1fd0b948b4f64e8134893e6e913a06af2800..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/normalization/device_normalization_f32_instance.cpp
+++ /dev/null
@@ -1,60 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
-#include "ck/utility/data_type.hpp"
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F32 = float;
-
-using Pass = ck::tensor_operation::element_wise::PassThrough;
-
-template <typename OutElementwise, index_t Rank, index_t Reduce>
-using device_layernorm_f32_instances = std::tuple<
-    // clang-format off
-        // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 1, 1, 1, 1, 1, 1>, // fallback kernel
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 2, 1, 2, 1, 2, 2>, // fallback kernel
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 8, 32, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 4, 64, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 2, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>
-    // clang-format on
-    >;
-
-void add_device_normalization_rank_2_1_f32_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 2, 1>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_layernorm_f32_instances<Pass, 2, 1>{});
-}
-
-void add_device_normalization_rank_4_3_f32_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 4, 3>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_layernorm_f32_instances<Pass, 4, 3>{});
-}
-
-void add_device_normalization_rank_5_3_f32_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 5, 3>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_layernorm_f32_instances<Pass, 5, 3>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..b0684962f9ef649d1f90a83d40fb2b76c085c549
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
@@ -0,0 +1,122 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f16_instances =
+    // clang-format off
+    std::tuple <
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8>
+        // clang-format on
+        >;
+
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f16_generic_instance = std::tuple<
+    // clang-format off
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+    // clang-format on
+    >;
+
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f32_instances = std::tuple<
+    // clang-format off
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+    // clang-format on
+    >;
+
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f32_generic_instance = std::tuple<
+    // clang-format off
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+    // clang-format on
+    >;
+
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f16_f32_f32_f16_instances = std::tuple<
+    // clang-format off
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+    // clang-format on
+    >;
+
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f16_f32_f32_f16_generic_instance = std::tuple<
+    // clang-format off
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0d0f896c8d932384fbe8e3e84a7f1d0e6b2c388d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
@@ -0,0 +1,10 @@
+add_instance_library(device_pool_fwd_instance
+    device_avg_pool2d_fwd_nhwc_f16_instance.cpp
+    device_avg_pool2d_fwd_nhwc_f32_instance.cpp
+    device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+    device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+    device_max_pool2d_fwd_nhwc_f16_instance.cpp
+    device_max_pool2d_fwd_nhwc_f32_instance.cpp
+    device_max_pool3d_fwd_ndhwc_f16_instance.cpp
+    device_max_pool3d_fwd_ndhwc_f32_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..508ad3873b65e0cf7109adb2ffc1f7d7666149eb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ada96a93a2dfe1c71e83d3f14fdf2bb4b4647da0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..62bcad992ae1b8d77bfea25f673191979e1aeab8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..47896be911571933136b04e726d7273eb635254c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..35c8522d9f9aedb64c15338a3e8389102534262b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F16, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F16, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..75b7629f246c70c99bc8f021bc299142609bbd3d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..dbfc4acfdfbed1eb1c22a410e4d54edc915c39fe
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F16, ReduceOpId, false>{});
+}
+
+void add_device_pool3d_fwd_ndhwc_index_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F16, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..63b3e8df8ea3254ee7f5fc0cb4be2270a144a72f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool3d_fwd_ndhwc_index_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp b/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..8aa707885b2b785974ecef5491cae9ac7ec391d3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I32 = int32_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename ComputeDataType,
+          ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+using device_pool2d_fwd_nhwc_instances =
+    // clang-format off
+    std::tuple <
+        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 1, 1, 1>,
+        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 2, 1, 2>,
+        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 4, 1, 4>
+               // clang-format on
+               >;
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename ComputeDataType,
+          ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+using device_pool3d_fwd_ndhwc_instances =
+    // clang-format off
+    std::tuple <
+        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 1, 1, 1>,
+        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 2, 1, 2>,
+        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 4, 1, 4>
+               // clang-format on
+               >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
index 9f826afd68074d0e73aeb19ebfdc672aa26fcb04..db650dbfbdb41ad6a300c787a37fe6f8ea2719cc 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
@@ -1,6 +1,38 @@
+set(CONV2D_PERLAYER_QUANT_SRC
+    conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
+    conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
+)
+
+set(CONV2D_PERCHANNEL_QUANT_SRC
+    conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
+    conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
+)
+
+set(CONV2D_BIAS_PERLAYER_QUANT_SRC
+    conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
+    conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
+)
+
+set(CONV2D_BIAS_PERCHANNEL_QUANT_SRC
+    conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
+    conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
+)
+
+set(GEMM_QUANT_SRC
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+    gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+    gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+    gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+    gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+)
+
 add_instance_library(device_quantization_instance
-    device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
-    device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
-    device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
-    device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
+    ${CONV2D_PERLAYER_QUANT_SRC}
+    ${CONV2D_PERCHANNEL_QUANT_SRC}
+    ${CONV2D_BIAS_PERLAYER_QUANT_SRC}
+    ${CONV2D_BIAS_PERCHANNEL_QUANT_SRC}
+    ${GEMM_QUANT_SRC}
 )
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..711314985a7cd2ba708138df597e37d1e874709f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
@@ -0,0 +1,68 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Empty_Tuple = ck::Tuple<>;
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NHWGC       = ck::tensor_layout::convolution::NHWGC;
+using GKYXC       = ck::tensor_layout::convolution::GKYXC;
+using NHWGK       = ck::tensor_layout::convolution::NHWGK;
+using GK          = ck::tensor_layout::convolution::G_K;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Relu        = ck::tensor_operation::element_wise::Relu;
+using TanH        = ck::tensor_operation::element_wise::TanH;
+
+using GK_Tuple      = ck::Tuple<GK>;
+using GK_GK_Tuple   = ck::Tuple<GK, GK>;
+using I32_Tuple     = ck::Tuple<int32_t>;
+using F32_Tuple     = ck::Tuple<float>;
+using I32_F32_Tuple = ck::Tuple<int32_t, float>;
+
+// perlayer
+using Mul_Clamp      = ck::tensor_operation::element_wise::Activation_Mul_Clamp<PassThrough>;
+using Relu_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<Relu>;
+
+// bias + perlayer
+using Add_Mul_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<PassThrough>;
+using Add_Relu_Mul_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<Relu>;
+using Add_Mul_TanH_Mul_Clamp =
+    ck::tensor_operation::element_wise::Add_Mul_Activation_Mul_Clamp<TanH>;
+
+// perchannel
+using Mul2_Clamp      = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<PassThrough>;
+using Relu_Mul2_Clamp = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<Relu>;
+
+// bias + perchannel
+using Add_Mul2_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<PassThrough>;
+using Add_Relu_Mul2_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<Relu>;
+using Add_Mul2_TanH_Mul_Clamp =
+    ck::tensor_operation::element_wise::Add_Mul2_Activation_Mul_Clamp<TanH>;
+
+static constexpr ck::index_t NDimSpatial = 2;
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..39c4f82fefd4beb2a0dc3629f0757f7f01037495
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
@@ -0,0 +1,145 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_dl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul2_Clamp>>>& instances)
+{
+    // dl
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Mul2_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Mul2_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Mul2_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Relu_Mul2_Clamp>>>& instances)
+{
+    // dl
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Relu_Mul2_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Relu_Mul2_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Relu_Mul2_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul2_TanH_Mul_Clamp>>>& instances)
+{
+    // dl
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Mul2_TanH_Mul_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Mul2_TanH_Mul_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_F32_Tuple,
+                                                                           Add_Mul2_TanH_Mul_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..92e73eb2ee45895c4277eb46f2a5ec94116fc1e2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_dl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Mul_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Mul_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Mul_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Relu_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Relu_Mul_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Relu_Mul_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Relu_Mul_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul_TanH_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Mul_TanH_Mul_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Mul_TanH_Mul_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           I32_Tuple,
+                                                                           Add_Mul_TanH_Mul_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..2ec37c8413e7ca926996ed8165c7e23730e77769
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "conv2d_quantization_common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// clang-format off
+template <typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
+          typename DsDatatype,
+          typename OutElementOp,
+          ConvolutionForwardSpecialization ConvSpec,
+          index_t DstScalarPerVector>
+using device_grouped_conv2d_dl_int8_instances =
+    std::tuple<
+        // ###########################################|        NDim| InData| WeiData|   MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        // ###########################################|     Spatial|   Type|    Type|       Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        // ###########################################|            |       |        |           |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        // ###########################################|            |       |        |           |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< NDimSpatial, int8_t,  int8_t, DsDatatype,      int8_t,     int32_t, InLayout, WeiLayout,    DsLayout, OutLayout,  PassThrough,   PassThrough,  OutElementOp,       ConvSpec,          GemmSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5, DstScalarPerVector>
+    >;
+// clang-format on
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1d8b58fd18a49b777158c266459925a8ef67079b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
@@ -0,0 +1,98 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_dl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_dl_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Mul2_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           F32_Tuple,
+                                                                           Mul2_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           F32_Tuple,
+                                                                           Mul2_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           F32_Tuple,
+                                                                           Mul2_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Relu_Mul2_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           F32_Tuple,
+                                                                           Relu_Mul2_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           F32_Tuple,
+                                                                           Relu_Mul2_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
+                                                                           F32_Tuple,
+                                                                           Relu_Mul2_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..62826d0d2b3bb541cb70ed1fc7f70f6c55a431c1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
@@ -0,0 +1,98 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_dl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_dl_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
+                                                                           Empty_Tuple,
+                                                                           Mul_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
+                                                                           Empty_Tuple,
+                                                                           Mul_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
+                                                                           Empty_Tuple,
+                                                                           Mul_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+
+void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Relu_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
+                                                                           Empty_Tuple,
+                                                                           Relu_Mul_Clamp,
+                                                                           ConvFwdDefault,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
+                                                                           Empty_Tuple,
+                                                                           Relu_Mul_Clamp,
+                                                                           ConvFwd1x1P0,
+                                                                           4>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
+                                                                           Empty_Tuple,
+                                                                           Relu_Mul_Clamp,
+                                                                           ConvFwd1x1S1P0,
+                                                                           4>{});
+}
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..99877c32dba05ea6386bb2f94d888e7fe6d95279
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
@@ -0,0 +1,142 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_xdl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul2_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Mul2_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Mul2_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Mul2_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+
+void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Relu_Mul2_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Relu_Mul2_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Relu_Mul2_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Relu_Mul2_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+
+void add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul2_TanH_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Mul2_TanH_Mul_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Mul2_TanH_Mul_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_F32_Tuple,
+                                                                            Add_Mul2_TanH_Mul_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..50ccc69f458fc9a46bbe5e08510e3da1174a90cc
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_xdl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Mul_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Mul_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Mul_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+
+void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Relu_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Relu_Mul_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Relu_Mul_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Relu_Mul_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+
+void add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              I32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Add_Mul_TanH_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Mul_TanH_Mul_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Mul_TanH_Mul_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            I32_Tuple,
+                                                                            Add_Mul_TanH_Mul_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..caced6c950a7872714b76aadee100e3a524a03d8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "conv2d_quantization_common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// clang-format off
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename OutElementOp,
+          ConvolutionForwardSpecialization ConvSpec,
+          index_t DstScalarPerVector>
+using device_grouped_conv2d_xdl_int8_instances =
+    std::tuple <
+        //########################################|     NumDim|       A|       B|       Ds|       E|  AData|  BData| AccData| CShuffle|         Ds|  EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|     CBlockTransfer|
+        //########################################|    Spatial|  Layout|  Layout|   Layout|  Layout|   Type|   Type|    Type| DataType|   DataType|   Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|    ScalarPerVector|
+        //########################################|           |        |        |         |        |       |       |        |         |           |       |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|      _NWaveNPerXdl|
+        //########################################|           |        |        |         |        |       |       |        |         |           |       |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                   |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>
+    >;
+// clang-format on
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..526fe73463cba7ac977b597c2279488aaa9afe6b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
@@ -0,0 +1,98 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_xdl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Mul2_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            F32_Tuple,
+                                                                            Mul2_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            F32_Tuple,
+                                                                            Mul2_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            F32_Tuple,
+                                                                            Mul2_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+
+void add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              GK_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              F32_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Relu_Mul2_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            F32_Tuple,
+                                                                            Relu_Mul2_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            F32_Tuple,
+                                                                            Relu_Mul2_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            8>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
+                                                                            F32_Tuple,
+                                                                            Relu_Mul2_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            8>{});
+}
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d1d5a66bb949b391909426aa13ac4bc4d9e91be6
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
@@ -0,0 +1,98 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_conv2d_xdl_int8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
+                                                                            Empty_Tuple,
+                                                                            Mul_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            16>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
+                                                                            Empty_Tuple,
+                                                                            Mul_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            16>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
+                                                                            Empty_Tuple,
+                                                                            Mul_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            16>{});
+}
+
+void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              Relu_Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
+                                                                            Empty_Tuple,
+                                                                            Relu_Mul_Clamp,
+                                                                            ConvFwdDefault,
+                                                                            16>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
+                                                                            Empty_Tuple,
+                                                                            Relu_Mul_Clamp,
+                                                                            ConvFwd1x1P0,
+                                                                            16>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
+                                                                            Empty_Tuple,
+                                                                            Relu_Mul_Clamp,
+                                                                            ConvFwd1x1S1P0,
+                                                                            16>{});
+}
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
deleted file mode 100644
index e87e9875939e578dafcdaf973db9909f32894ec7..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
+++ /dev/null
@@ -1,74 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "device_conv2d_xdl_int8_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-void add_device_conv2d_bias_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              GK_GK_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Mul2_Clamp>>>& instances)
-{
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_GK_Tuple,
-                                                                     I32_F32_Tuple,
-                                                                     Add_Mul2_Clamp,
-                                                                     ConvFwdDefault>{});
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_GK_Tuple,
-                                                                     I32_F32_Tuple,
-                                                                     Add_Mul2_Clamp,
-                                                                     ConvFwd1x1P0>{});
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_GK_Tuple,
-                                                                     I32_F32_Tuple,
-                                                                     Add_Mul2_Clamp,
-                                                                     ConvFwd1x1S1P0>{});
-}
-
-void add_device_conv2d_bias_relu_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              GK_GK_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Relu_Mul2_Clamp>>>& instances)
-{
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_GK_Tuple,
-                                                                     I32_F32_Tuple,
-                                                                     Add_Relu_Mul2_Clamp,
-                                                                     ConvFwdDefault>{});
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_GK_Tuple,
-                                                                     I32_F32_Tuple,
-                                                                     Add_Relu_Mul2_Clamp,
-                                                                     ConvFwd1x1P0>{});
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_GK_Tuple,
-                                                                     I32_F32_Tuple,
-                                                                     Add_Relu_Mul2_Clamp,
-                                                                     ConvFwd1x1S1P0>{});
-}
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
deleted file mode 100644
index 06eed760149de52e26e351170d2cd310c4698e26..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
+++ /dev/null
@@ -1,68 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "device_conv2d_xdl_int8_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-void add_device_conv2d_bias_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Mul_Clamp>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, I32_Tuple, Add_Mul_Clamp, ConvFwdDefault>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, I32_Tuple, Add_Mul_Clamp, ConvFwd1x1P0>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, I32_Tuple, Add_Mul_Clamp, ConvFwd1x1S1P0>{});
-}
-
-void add_device_conv2d_bias_relu_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              I32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Add_Relu_Mul_Clamp>>>& instances)
-{
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_Tuple,
-                                                                     I32_Tuple,
-                                                                     Add_Relu_Mul_Clamp,
-                                                                     ConvFwdDefault>{});
-
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, I32_Tuple, Add_Relu_Mul_Clamp, ConvFwd1x1P0>{});
-
-    add_device_operation_instances(instances,
-                                   device_conv2d_int8_32Ds_instances<GK_Tuple,
-                                                                     I32_Tuple,
-                                                                     Add_Relu_Mul_Clamp,
-                                                                     ConvFwd1x1S1P0>{});
-}
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_int8_instance.hpp
deleted file mode 100644
index 6904e269f94af8c53298f99c62f205a1ca1cfcf9..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_int8_instance.hpp
+++ /dev/null
@@ -1,111 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using Empty_Tuple = ck::Tuple<>;
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC       = ck::tensor_layout::convolution::GNHWC;
-using GKYXC       = ck::tensor_layout::convolution::GKYXC;
-using GNHWK       = ck::tensor_layout::convolution::GNHWK;
-using GK          = ck::tensor_layout::convolution::G_K;
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Relu        = ck::tensor_operation::element_wise::Relu;
-
-using GK_Tuple      = ck::Tuple<GK>;
-using GK_GK_Tuple   = ck::Tuple<GK, GK>;
-using I32_Tuple     = ck::Tuple<int32_t>;
-using F32_Tuple     = ck::Tuple<float>;
-using I32_F32_Tuple = ck::Tuple<int32_t, float>;
-
-using Mul_Clamp      = ck::tensor_operation::element_wise::Activation_Mul_Clamp<PassThrough>;
-using Relu_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<Relu>;
-
-using Add_Mul_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<PassThrough>;
-using Add_Relu_Mul_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<Relu>;
-
-using Mul2_Clamp      = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<PassThrough>;
-using Relu_Mul2_Clamp = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<Relu>;
-
-using Add_Mul2_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<PassThrough>;
-using Add_Relu_Mul2_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<Relu>;
-
-static constexpr ck::index_t NDimSpatial = 2;
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-template <typename DsLayout,
-          typename DsDatatype,
-          typename OutElementOp,
-          ConvolutionForwardSpecialization ConvSpec>
-// clang-format off
-using device_conv2d_int8_instances =
-    std::tuple <
-        //########################################|  NumDim|      A|      B|       Ds|      E|  AData|  BData| AccData| CShuffle|         Ds|  EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|   Layout| Layout|   Type|   Type|    Type| DataType|   DataType|   Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |         |       |       |       |        |         |           |       |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |         |       |       |       |        |         |           |       |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>,               16>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>,               16>
-    >;
-// clang-format on
-
-// for conv + multiple of 32 bit Ds. bit of Ds will affect the ScalarPerVector of C
-template <typename DsLayout,
-          typename DsDatatype,
-          typename OutElementOp,
-          ConvolutionForwardSpecialization ConvSpec>
-// clang-format off
-using device_conv2d_int8_32Ds_instances =
-    std::tuple <
-        //########################################|  NumDim|      A|      B|       Ds|      E|  AData|  BData| AccData| CShuffle|         Ds|  EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|   Layout| Layout|   Type|   Type|    Type| DataType|   DataType|   Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |         |       |       |       |        |         |           |       |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |         |       |       |       |        |         |           |       |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>,               8>
-    >;
-// clang-format on
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
deleted file mode 100644
index 5f1aa0c5c77afac489e571b7b5dad79e75c37efa..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
+++ /dev/null
@@ -1,62 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "device_conv2d_xdl_int8_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-void add_device_conv2d_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Mul2_Clamp>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, F32_Tuple, Mul2_Clamp, ConvFwdDefault>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, F32_Tuple, Mul2_Clamp, ConvFwd1x1P0>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, F32_Tuple, Mul2_Clamp, ConvFwd1x1S1P0>{});
-}
-
-void add_device_conv2d_relu_perchannel_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              GK_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              F32_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Relu_Mul2_Clamp>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, F32_Tuple, Relu_Mul2_Clamp, ConvFwdDefault>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, F32_Tuple, Relu_Mul2_Clamp, ConvFwd1x1P0>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_32Ds_instances<GK_Tuple, F32_Tuple, Relu_Mul2_Clamp, ConvFwd1x1S1P0>{});
-}
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
deleted file mode 100644
index 83435d8119a7beae3bc6964f3ce8514bf7e8000b..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/quantization/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
+++ /dev/null
@@ -1,62 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "device_conv2d_xdl_int8_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-void add_device_conv2d_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Mul_Clamp>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_instances<Empty_Tuple, Empty_Tuple, Mul_Clamp, ConvFwdDefault>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_instances<Empty_Tuple, Empty_Tuple, Mul_Clamp, ConvFwd1x1P0>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_instances<Empty_Tuple, Empty_Tuple, Mul_Clamp, ConvFwd1x1S1P0>{});
-}
-
-void add_device_conv2d_relu_perlayer_quantization_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              Relu_Mul_Clamp>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_instances<Empty_Tuple, Empty_Tuple, Relu_Mul_Clamp, ConvFwdDefault>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_instances<Empty_Tuple, Empty_Tuple, Relu_Mul_Clamp, ConvFwd1x1P0>{});
-    add_device_operation_instances(
-        instances,
-        device_conv2d_int8_instances<Empty_Tuple, Empty_Tuple, Relu_Mul_Clamp, ConvFwd1x1S1P0>{});
-}
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0ec498a441d26c37cb5e449213f9b1048c36c9c5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gemm_quantization_common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <typename OutElementOp>
+using device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances = std::tuple<
+    // clang-format off
+        //####################|      A|      B|          Ds|      E|  AData|  BData| AccData|      DsData|   EData|           A|           B|           CDE|           GEMM| Block|  MPer|  NPer|  KPer| K1|   M1Per|  N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        //####################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM| Thread| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        //####################|       |       |            |       |       |       |        |            |        |   Operation|   Operation|     Operation|               |      |      |      |      |   |        |       |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        //####################|       |       |            |       |       |       |        |            |        |            |            |              |               |      |      |      |      |   |        |       |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGemmMultipleD_Dl<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,   256,   128,   128,    16,  4,       4,      4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 4>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+    // clang-format on
+    >;
+
+template <typename OutElementOp>
+using device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instances = std::tuple<
+    // clang-format off
+        //####################|      A|      B|          Ds|      E|  AData|  BData| AccData|      DsData|   EData|           A|           B|           CDE|           GEMM| Block|  MPer|  NPer|  KPer| K1|   M1Per|  N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        //####################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM| Thread| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        //####################|       |       |            |       |       |       |        |            |        |   Operation|   Operation|     Operation|               |      |      |      |      |   |        |       |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        //####################|       |       |            |       |       |       |        |            |        |            |            |              |               |      |      |      |      |   |        |       |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGemmMultipleD_Dl<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,   256,   128,   128,    16,  4,      4,       4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+    // clang-format on
+    >;
+
+template <typename OutElementOp>
+using device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //####################|      A|      B|          Ds|      E|  AData|  BData| AccData|      DsData|   EData|           A|           B|           CDE|           GEMM| Block|  MPer|  NPer|  KPer| K1|   M1Per|  N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        //####################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM| Thread| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        //####################|       |       |            |       |       |       |        |            |        |   Operation|   Operation|     Operation|               |      |      |      |      |   |        |       |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        //####################|       |       |            |       |       |       |        |            |        |            |            |              |               |      |      |      |      |   |        |       |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGemmMultipleD_Dl<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,   256,   128,   128,    16,  4,       4,      4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+    // clang-format on
+    >;
+
+template <typename OutElementOp>
+using device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //####################|      A|      B|          Ds|      E|  AData|  BData| AccData|      DsData|   EData|           A|           B|           CDE|           GEMM| Block|  MPer|  NPer|  KPer| K1|   M1Per|  N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        //####################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM| Thread| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        //####################|       |       |            |       |       |       |        |            |        |   Operation|   Operation|     Operation|               |      |      |      |      |   |        |       |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        //####################|       |       |            |       |       |       |        |            |        |            |            |              |               |      |      |      |      |   |        |       |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGemmMultipleD_Dl<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,   256,   128,   128,    16,  4,       4,      4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..4c02f8179caf2a7bf5ee6a46111f0b9d610e7667
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Col, Row, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances<Mul_Clamp>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..2c635e96d228fb4283fe518dbf010c66fc9a4272
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Col, Col, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instances<Mul_Clamp>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c5a52f2cf25137fbbfac20d6451db1c67c1ea660
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Row, Row, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instances<Mul_Clamp>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..fe4a0b4d68b4ed659aef0e6ab23d4a443cdfde72
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Row, Col, Row]
+void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances<Mul_Clamp>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..d1b18ac4929bb114ab870a735b6b182e7a4e1343
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp
@@ -0,0 +1,116 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gemm_quantization_common.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <typename OutElementOp, LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|      DsData|   EData|           A|           B|           CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     LoopScheduler|     Pipeline|
+        //##############################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |   Operation|   Operation|     Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |            |            |              |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                  |             |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   256,   128,    64,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   256,    64,   4,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,   128,    64,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   128,    64,   4,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    64,    64,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    64,   128,    64,   4,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,    64,    64,   4,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,    64,   128,    64,   4,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+template <typename OutElementOp, LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|      DsData|   EData|           A|           B|           CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     LoopScheduler|     Pipeline|
+        //##############################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |   Operation|   Operation|     Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |            |            |              |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                  |             |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   256,   128,    64,   4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   256,    64,   4,  16,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,   128,    64,   4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   128,    64,   4,  16,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    64,    64,   4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    64,   128,    64,   4,  16,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,    64,    64,   4,  16,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,    64,   128,    64,   4,  16,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+template <typename OutElementOp, LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|      DsData|   EData|           A|           B|           CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     LoopScheduler|     Pipeline|
+        //##############################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |   Operation|   Operation|     Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |            |            |              |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                  |             |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   256,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   256,    64,  16,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    64,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    64,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,    64,    64,  16,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,    64,   128,    64,  16,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+template <typename OutElementOp, LoopScheduler GemmLoopScheduler, PipelineVersion GemmPipeline>
+using device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|      DsData|   EData|           A|           B|           CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     LoopScheduler|     Pipeline|
+        //##############################| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|        Type|    Type| Elementwise| Elementwise|   Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |   Operation|   Operation|     Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                  |             |
+        //##############################|       |       |            |       |       |       |        |         |            |        |            |            |              |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                  |             |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Empty_Tuple,    Row, int8_t, int8_t, int32_t,  int32_t, Empty_Tuple,  int8_t, PassThrough, PassThrough,  OutElementOp,     MNKPadding,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,              16, GemmLoopScheduler, GemmPipeline>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e3163c32ed1cb9704e00f5d46e82cfacb187aa60
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Col, Row, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Interwave,
+                                                                           PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v2>{});
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f6a0163aba346bc731f3241f0426306e5eeb43f3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Col, Col, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Interwave,
+                                                                           PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v2>{});
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0cfffbed70fdc95b6161c06a634bf379cc7c2651
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Row, Row, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Interwave,
+                                                                           PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v2>{});
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..289898257b76ab48c44e06259dd067096f8489cd
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Layout(A, B, C) = [Row, Col, Row]
+void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Empty_Tuple,
+                                                    Row,
+                                                    int8_t,
+                                                    int8_t,
+                                                    Empty_Tuple,
+                                                    int8_t,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Mul_Clamp>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v1>{});
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Interwave,
+                                                                           PipelineVersion::v1>{});
+#endif
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES
+    add_device_operation_instances(
+        instances,
+        device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances<Mul_Clamp,
+                                                                           LoopScheduler::Default,
+                                                                           PipelineVersion::v2>{});
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/gemm_quantization_common.hpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/gemm_quantization_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..e7c2500fef2fac9e90639301c60592dcb8978ccd
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/gemm_quantization_common.hpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using Empty_Tuple   = ck::Tuple<>;
+using Row_Row_Tuple = ck::Tuple<Row, Row>;
+using Col_Col_Tuple = ck::Tuple<Col, Col>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Relu        = ck::tensor_operation::element_wise::Relu;
+
+using Mul_Clamp      = ck::tensor_operation::element_wise::Activation_Mul_Clamp<PassThrough>;
+using Relu_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<Relu>;
+
+using Add_Mul_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<PassThrough>;
+using Add_Relu_Mul_Clamp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<Relu>;
+
+static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp
index 1909183a55c72fec95250d29a34f80c0d36b1758..4771b4aa5de2ccfb5b7f9a6205f64594d37bdabd 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp
index ec302010219b84d6109bd7eae209d6f9fef1ff0d..f684dabc1fcd60bd6522a9865f18ac39dcdc3d0e 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp
index 89f3e582802f0d030daf734e11fc8811640cafb1..86e2ca7c47cb8f2a318c745b19e0fcec058c7318 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp
index f1bdd1927b1fe035e1393fb5ec09856db0cecaa2..97d09ff834fc46a65b8f25182238349f70bce818 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp
index 58e9c5622951dd38cf49eecb013a04abfb7f610a..87187f493acf8b8247e8877b369edfc106a6aaa1 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp
index e5012c651aa28e7e4ba78c8caddcbd59c602c8c3..d3f5cd8656940f1adad49d0fd38725a5e04167ea 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp
index 0970cb9d7c2769ed87d96b7320fef43992e4f6c0..fb4a50d5e3539c3c2db9e799ea2c824a59c90c7c 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp
index 6ee179a5117a21324c1a105fbf67f5df45da6696..f74d22f7acadb2ab97a665adfb6eb079ed34cb64 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp
index e53b403065490e0148c8ce47107dd892d13bd54a..0982455add502da7eb5f9365f4b128673846db82 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp
index cab5738fbae05b7e2ee68a1e2ca937c94601f2ea..c6f1c3a6338115a42efc02311228af8b488c4a8d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp
index 7d2a4fad2a85805b177575477c6ea3c69a93effe..db9f52a13fe436252d57644b9d76db41170b042b 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp
index e08b64f8b37d8b10d799aaeceeb19a7bccd9df4a..1410f51bea100f3aa56b369d8442d3c81fc79ebe 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp
index 89cabf3762355c7c235e36e3e8701acdb7c95c77..707a655703503001118da5d2e28e53f24ecc135c 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp
index 1e602c121d0c9d502cd01dd862a39ed695854a22..17f45c3327c63296291aef9132461f5bb56074b2 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp
index 489b4bc452f6017bee956517ebd9b2d2047ab182..d3753d01c4e65dc2344ff4d8199bd81d674f5c96 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp
index 04e2c5b164fb5a803336c90154f2e938c421e29c..83fad7826a070ef91abcc33c68b1230b410fb8f5 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp
index 5c0e53604851ab08808e23e75ca57c9887053725..c627f686363a9a12e251797b45060d56d2947d33 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp
index 899dfcd37c1b39b3d62b1c0a623f4edbab68a4c4..09686d98ffcacb07f5f9244c196daeec993780dc 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp
index 5624337a47735470d4a78c1280afda20067be8de..44519032f31cba47f9606aee66dce5b989685931 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp
index 2f3067ce291465a71656eed576986877b7496de7..9e14d80e21ae1956595eca87ab6ab859c42c52f6 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp
index 2648e7d59dba49520a1e87870ebce49b4197f9de..ab217f1160481e0155cc3f0e6e62dc21f1f70fe2 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp
index f67ae2ee7c6c5b74ef6509a467e6864e16d05fa4..3f611da7e0da0d7f61b3af38f18ac29b8dc5b2a0 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp
index 6f8e07851dfcfdb8f4aa51d4aa030b65d527d9fb..8695db4004081ee0197eebd623cf539286236d3e 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp
index 69fecf72f510b6a5246842ba15f886c6d9d842b6..8d15fda7869d85daa457ff5cfb69bfcbb719cf9b 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp
index 129a4f0f0e87a4de66d072596c0cb7ff23af3caa..c161858b1f161694485d5531485995490949112d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp
index 21babc4aa6315d59705c55de78410e79a9411a97..e6b87fdead361de49fbb28e5a709b2d5fea812b2 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp
index b85b3e2b68e769367cd955d086249b2bdd27f79c..c2e4acc8d9338824df7811d2434474ceddf5ec6d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp
index 24a8293b5dda85cd8876c28373710b4398f5af0b..e5a5e07e176b68ed8eca614a52a432efff1ca9d7 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp
index 73e60fa959ee4654958a93a230e584e02b4e3329..22c62033454a9525f87fc7b19f8ab0f6186a3cda 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp
index 72e649d8971ff220caeca22b63102da0417039c4..1890fb78cd7c7fd387f6f9d3290009b54ca3ba96 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp
index a7e053a06564dc90acbc5540e998b8063496c92c..dc5caf297bc971950c2424496ec1c2a2fe2c2a91 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp
index 0e3abd35b46f9525fdb026a7bec6480555a70c43..02bb0c3b1d9d5e8db84a0338718c66a4cc7e66c9 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp
index 4b32456074f09313864fb3f1829e046478fab27d..0b0a40303805ccfb4d9ea93e4b1295cc7b2edb6e 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp
index 3298587a42f30db880f0180605e47d50189fc33b..fcf95aab7a37aa9c79ef6e22717afebb24052406 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp
index 729d4fd6e19b470fa47f8a33ca5336c20441ca30..5151d32a58e9c321985faf6f8cb7508e5cc4ad5f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp
index e3e36e312ba1607bd0707bac69cc3ab2fa137adc..5463cc2b4c2e5762f08db5593bfc51e204a41713 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp
index e7580e7d7dc3b030f6bb5f27b5c68a8b04be1ce8..a15e99fe0f57a2aab28811660dd910645cef482a 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp
index 1e6feb0071fe6b2798da98b231ac2168efcbb067..ee7a4dd4cef85adf56094750c09558b3dea1ec5a 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp
index 669c4d34ca751ebb7550a4ee447a03f7bc582f00..58f7295a835ac90d6bc13125f2dc290544497bc0 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp
index 335a5474ce8985f17b5f8d4444b5af1020a150db..43f2370a2c37cbe181ac99b81c51313f1b89d3b3 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp
index e95e8391a27978f27709aed71f6fb53ac3b324d6..eaf84a6de1b7ba55446a2994ff5cdf9eb939e625 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp
index 25498158a2aa715c0cc73688b7ea6ef96bbf60a1..76d63c9a96372e9c3cd97de281d4e841d92ee71a 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp
index 7262b8a5ba6342e4bc1160aa3edc1cdf1856ba6e..eb1d389c398fe609b60d16c9202ef2f117cd43ec 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp
index c526a74f1a9a8630c8fecdd2e5da440d296621e3..61225df05dd1d0e444c76fdf3c0a318b80d17380 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp
index 4c7252e742d2ba23272a9378337522c6b09eb703..aef231e17ac12f2b6380213ce810831da52c7ad1 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp
index 618900a7d75c6e2c478abc6cad9859660deab702..6d2939e3e0e3ab7e196c390b4758bd0ab0345ba2 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp
index ce747cbc76473a91983728c0838f4abd23237fca..ff0ddd773b75603ca245bd5ca3821d5443db00d9 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp
index 06f622b9e69a91e0f58a587497e34a56f92d287f..1367022065226226457941e729b71c0d1b66b387 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp
index 708eb58d4041edc70c0b245160637d39dbbc30d3..f0f93019189aa751fbc39dcafed73cc42fd0f85c 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp
index c8a62fa1496a5a981e4f82661c403893286244cf..db9f1566008d81c0c5d5e09f2b73c13d7c77fdad 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp
index ce2092153cc9516895a1315718f88bac806f6775..e5644faf291740876ee4daeaf66fc0e43e20b0ce 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp
index 29251a8b9a6baebec2c104ef920e814cea82ecb0..350259f5f0c1ebc258bd202f182e54d600b619cd 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp
index 734fa9fd3e1494f6447a54badae8d9c87dcce0db..89cae5527342997a0db882b1c438a8dc117e972f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp
index d7a0e2bfe8932a4566640c014c1f3db9b051a8a6..b0f82ab45350b5bb71e15f609dc0e01264df8728 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp
index 8b97f3008b8b0eff0cee4ff41e73d55e5cd66e85..fedd0f890bd8987a1bf1b9b7f29d99c5293676f2 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp
index 53d01e38d60c93ab989bc5d2be940d7af3c19081..64ec1ce1574a3608be4db2a88b02d26c1f1c0835 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp
index 125d054f3dcf21f668110ae0c792138f335d0420..156f182f55a1e515e337b7a3cd20f124f618fcec 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp
index fb86a2bbe443b318d04889c9b967244926cc6e91..22b278790d970dbb4cb68741f653606cbd922abb 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp
index 49af08390ad7de3dfc94bde69faaa1f0f374d294..f3070d297587f8510ee21b553d7299a9cd30d745 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp
index 30cc1b13ecab4c224d7e01f734d3079e7517811a..e17476ba75111134f67457df836415323c93a2ce 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp
index 24f8a9ba5cf22f30a620e3406fe923b20f86a65d..2387725b9f74b543e1db6ca3de372838a9badaea 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp
index a26702f053cc98ab67ddb50b15743bc312f82dd4..8be94118c4864ec38087204d87f1fbb83b85ec43 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp
index 34fe32628fd5b4e83205204c86599c2ef8c07e64..9ff01615c9c83a51ac995cd5a7a6eb840c6fafa5 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp
index 74b15eddbacb01fa73d70b3708853c51d9c58600..8a99c7c95879554615a61e46a998ef213c9a5cdb 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp
index 65762492f765b103682cfc4f9b30b23580fa5c5e..78909c8dbf5f35e0241a1802c4e589085b4e7ead 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp
index 5e74295a0dc60c893cde4eb6fff9ac7704365598..7f094c28b756e020a47b1a78454f839ac9fb540e 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp
index 6fdea6cc4df1e6fd112382241cfa491ae2bf2cfc..4dfb477c1e74f4a3863df56fddbe815ea5d8a321 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp
index 317d573dac5352df541f559cfb6979a9be440efb..e4928336069cc19a6d161d035407ed8a63deefaf 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp
index 29f95ebcc7dfb3dcdd47c998b83b07f4d24a1b77..de8dce357bddc3470d8d17142b11f0cac7c6a507 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp
index aa9f47cbc44fe8e59218425089784d92b14e31b9..0225e75cb09a770c94597dd0caffee13f05b1b27 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp
index 54a9dd1ab7e333326c841d42b6852abe5a82ec35..6d04d2128f651c45b9abd4c2265c89c60495d45d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp
index 4ef5717b5e18edf3283d4da4f5851cace4c59005..046643abe1a6cb951204d3e624f89f678be71176 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp
index 140a3c197b0606357da23e6a065aa86143060ca0..a61302c001f7ecbcad117f5f8df38cc9ec9c3c14 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp
index 317b4ad39c0b014b0f04f88920bde86dc2af833b..7619e12a8e6bfe37e09abccb2f2672b9458cc934 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
index fc13261a6a74c49594240969cdae03ee4f5e5b7c..202ad12b9727cae983f11b800abf26cffad48926 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
@@ -1,13 +1,4 @@
 add_instance_library(device_softmax_instance
-    device_softmax_i8_i8_instance.cpp
-    device_softmax_i8_i8_instance_rank3_reduce1.cpp
-    device_softmax_i8_i8_instance_rank3_reduce2.cpp
-    device_softmax_i8_i8_instance_rank3_reduce3.cpp
-    device_softmax_i8_i8_instance_rank4_reduce1.cpp
-    device_softmax_i8_i8_instance_rank4_reduce2.cpp
-    device_softmax_i8_i8_instance_rank4_reduce3.cpp
-    device_softmax_i8_i8_instance_rank4_reduce4.cpp
-    device_softmax_f16_f16_instance.cpp
     device_softmax_f16_f16_instance_rank3_reduce1.cpp
     device_softmax_f16_f16_instance_rank3_reduce2.cpp
     device_softmax_f16_f16_instance_rank3_reduce3.cpp
@@ -15,7 +6,6 @@ add_instance_library(device_softmax_instance
     device_softmax_f16_f16_instance_rank4_reduce2.cpp
     device_softmax_f16_f16_instance_rank4_reduce3.cpp
     device_softmax_f16_f16_instance_rank4_reduce4.cpp
-    device_softmax_f32_f32_instance.cpp
     device_softmax_f32_f32_instance_rank3_reduce1.cpp
     device_softmax_f32_f32_instance_rank3_reduce2.cpp
     device_softmax_f32_f32_instance_rank3_reduce3.cpp
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
deleted file mode 100644
index 14d2764529c826587b13661e7a7cab11e8d6ea99..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f16_f16_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances)
-{
-    add_device_softmax_f16_f16_rank3_reduce1_instances(instances);
-    add_device_softmax_f16_f16_rank3_reduce2_instances(instances);
-    add_device_softmax_f16_f16_rank3_reduce3_instances(instances);
-}
-
-void add_device_softmax_f16_f16_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances)
-{
-    add_device_softmax_f16_f16_rank4_reduce1_instances(instances);
-    add_device_softmax_f16_f16_rank4_reduce2_instances(instances);
-    add_device_softmax_f16_f16_rank4_reduce3_instances(instances);
-    add_device_softmax_f16_f16_rank4_reduce4_instances(instances);
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
index fa334b997c27af20f05c42335bde28f5f32f819a..36867d993f9163bd7bdb8e81eb0ed51920ad2458 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f16_f16_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<3, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<3, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
index 1c9d37d8483b7f249a0b1c05551c70acdb46db03..373f33ad59716bcc2ddeb628a92dd0d46f7ed0aa 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f16_f16_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<3, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<3, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
index 5fbdab5055edb0f124bbc6541fcc705a12b71d63..d26b92b4f4998919d046401fb5f813594def112c 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f16_f16_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<3, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<3, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
index 7dd8640b187a2b793d5ad6931ab329ae024ca432..bbb735b6fe564e586cc269eb030d2dc38d4f7cd1 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
index b32fe6838f83440e98f524c59af9a383760dfe7d..92dbe6776039a9d8e3fd2d358abac54d33fe4537 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
index c05048ec567bd6471fca14c41d062eaf8c0c86a5..354cda85d757cb04d2b803ffec32e3a6c36f1856 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
index 6a235708bd426c7270f16246fa058eae45cdd056..edb5e42c103badf9fd65f482577db9ccc2ee08ab 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 4>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 4>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 4>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 4>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
deleted file mode 100644
index e5bec5e2639d234b3944f679e99e988f20b2c383..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f32_f32_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances)
-{
-    add_device_softmax_f32_f32_rank3_reduce1_instances(instances);
-    add_device_softmax_f32_f32_rank3_reduce2_instances(instances);
-    add_device_softmax_f32_f32_rank3_reduce3_instances(instances);
-}
-
-void add_device_softmax_f32_f32_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances)
-{
-    add_device_softmax_f32_f32_rank4_reduce1_instances(instances);
-    add_device_softmax_f32_f32_rank4_reduce2_instances(instances);
-    add_device_softmax_f32_f32_rank4_reduce3_instances(instances);
-    add_device_softmax_f32_f32_rank4_reduce4_instances(instances);
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
index 57d3f184a6635f4febe632054c9e39798ef0cc82..566be8fc22c4a5b8a5a6bb17ddebf012bf39d1d4 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f32_f32_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<3, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<3, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
index fae3a4dd6662d8920b1b88c8489e9a6f00f28ed0..f9c76e3116cd412c4efaca27595c5ebedf205943 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f32_f32_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<3, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<3, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
index b6fb70e8e2a6aa097f2a76865f690812dffaffdd..541e0d71a939c5964e560d76771986bdbc1cfc03 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f32_f32_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<3, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<3, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
index 33c7b6f35f351f755a50f48f92d391fd47f40163..95a38df2834b1794384bfdab57087c051a8fc482 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
index c22aa574b1f984d5a4f8c3276c9fb11b74002680..a29b88891d45b7f967629d956b69bc25c371ef88 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
index 55f3d2bd207a0a160f00415592f0106f41be5cee..0da46ea1b47489e9eb9fff991c7c5730e4b5338c 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
index fb0bcf5ee8a2dd6fd5ed4f77c0d8f4be5c5fcacd..fa217dc3f5b341112645c95df44826f19d2105c4 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 4>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 4>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 4>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 4>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
deleted file mode 100644
index 608cfcf8380be29e24d5e4b3aadd7573285a5224..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances)
-{
-    add_device_softmax_i8_i8_rank3_reduce1_instances(instances);
-    add_device_softmax_i8_i8_rank3_reduce2_instances(instances);
-    add_device_softmax_i8_i8_rank3_reduce3_instances(instances);
-}
-
-void add_device_softmax_i8_i8_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances)
-{
-    add_device_softmax_i8_i8_rank4_reduce1_instances(instances);
-    add_device_softmax_i8_i8_rank4_reduce2_instances(instances);
-    add_device_softmax_i8_i8_rank4_reduce3_instances(instances);
-    add_device_softmax_i8_i8_rank4_reduce4_instances(instances);
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
deleted file mode 100644
index 15552dbae5d501c506f4fd14b439a80664a95b66..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 3;
-
-void add_device_softmax_i8_i8_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 1>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
deleted file mode 100644
index 67674028860b471888bc87a59e1b7ae751c64eec..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 3;
-
-void add_device_softmax_i8_i8_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 2>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
deleted file mode 100644
index 4b33da93c2e1330485a55f8ede6654ab2a03fbb1..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 3;
-
-void add_device_softmax_i8_i8_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 3>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
deleted file mode 100644
index fe3b823e889267f69001e15066bbef201afa9813..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 1>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
deleted file mode 100644
index 8ecdf87d9fec061094f83c3beba01497dcb8e5b8..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 2>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
deleted file mode 100644
index 3563135204085ff8e44d1378f9c4d9ffd99e7b68..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 3>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp
deleted file mode 100644
index aa21a0bf8a863f433665e01e8229bd179948e1db..0000000000000000000000000000000000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 4>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/utility/convolution_parameter.cpp b/library/src/utility/convolution_parameter.cpp
index c8712d20939d5f9f1b7b981645e77420c7a5c607..57cedd60199e1c3aa291d44c404d56e89f444e9a 100644
--- a/library/src/utility/convolution_parameter.cpp
+++ b/library/src/utility/convolution_parameter.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/host_utility/io.hpp"
 
diff --git a/library/src/utility/device_memory.cpp b/library/src/utility/device_memory.cpp
index 90f943313b0961bb96d6394e855809c62d050559..11166783e8e542f565bfc1171ae40d879a005268 100644
--- a/library/src/utility/device_memory.cpp
+++ b/library/src/utility/device_memory.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/host_utility/hip_check_error.hpp"
 
diff --git a/library/src/utility/host_tensor.cpp b/library/src/utility/host_tensor.cpp
index e34fbc8f345b8ca6dd5b34988a1839c6c59e61bd..7211552641195d68e42ff606ddc1b2c7aab7ba84 100644
--- a/library/src/utility/host_tensor.cpp
+++ b/library/src/utility/host_tensor.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cassert>
 
diff --git a/profiler/README.md b/profiler/README.md
index bfd6a3a53be6ad8c84bb35c97d09f6125145591f..a18ebc226547cb1c58aea2900629bc67b9e8ee46 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -46,3 +46,141 @@ out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
 ....
 Best Perf: 1.42509 ms, 102.988 TFlops, 234.086 GB/s
 ```
+
+## Profile contraction kernels
+```bash
+#arg1: tensor operation (contraction_bilinear=CONTRACTION+Bilinear)
+#arg2: data type (0: fp32; 1: f64)\n"
+#arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1])
+#arg4: verification (0: no; 1: yes)
+#arg5: initialization (0: no init; 1: integer value; 2: decimal value)
+#arg6: print tensor value (0: no; 1: yes)
+#arg7: time kernel (0: no, 1: yes)
+#arg8 and arg9: alpha and beta
+#arg10 to 15: M0, M1, N0, N1, K0, K1
+#arg16 to 31: Strides for A, B, D and E (skip for default)
+
+################                   op  datatype  layout  verify  init  log  time  alpha  beta  M0  M1  N0  N1  K0  K1
+./bin/ckProfiler contraction_bilinear         0       1       0     0    0     1    1.0   1.0 128 128 128 128 128 128
+```
+
+Result (MI100)
+```bash
+a_m_k: dim 4, lengths {128, 128, 128, 128}, strides {2097152, 16384, 128, 1}
+b_k_n: dim 4, lengths {128, 128, 128, 128}, strides {128, 1, 2097152, 16384}
+d_m_n: dim 4, lengths {128, 128, 128, 128}, strides {2097152, 16384, 128, 1}
+e_m_n: dim 4, lengths {128, 128, 128, 128}, strides {2097152, 16384, 128, 1}
+....
+Best Perf: 211.405 ms, 41.6077 TFlops, 15.2372 GB/s
+```
+
+## Profile batched gemm multiple D kernels
+```bash
+#arg1: tensor operation (batched_gemm_multi_d=Batched GEMM multi D);
+#arg2: data type (0: fp16; 1: int8)
+#arg3: matrix layout (0: A[g, m, k] * B[g, k, n] = C[g, m, n];
+#                     1: A[g, m, k] * B[g, n, k] = C[g, m, n];
+#                     2: A[g, k, m] * B[g, k, n] = C[g, m, n];
+#                     3: A[g, k, m] * B[g, n, k] = C[g, m, n])
+#arg4: verification (0: no; 1: yes)
+#arg5: initialization (0: no init; 1: integer value; 2: decimal value)
+#arg6: print tensor value (0: no; 1: yes)
+#arg7: time kernel (0=n0, 1=yes)
+#arg8 to 17: M, N, K, StrideA, StrideB, StrideC, BatchStrideA, BatchStrideB, BatchStrideC, BatchCount
+
+################                   op  datatype  layout  verify  init  log  time    M    N    K StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
+./bin/ckProfiler batched_gemm_multi_d         0       1       0     0    0     1 4096 4096 4096    4096    4096    4096     16777216     16777216     16777216         16
+```
+
+Result (Radeon RX 6800 XT)
+```bash
+arg.a_grid_desc_k0_m0_m1_k1_{2048, 4096, 2}
+arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
+arg.e_grid_desc_m_n_{ 4096, 4096}
+....
+Best Perf: 58.0306 ms, 37.8942 TFlops, 27.7545 GB/s
+## Profile grouped convolution backward data kernels
+```bash
+# arg1: tensor operation (grouped_conv_bwd_data: Grouped Convolution Backward Data)
+# arg2: data type (0: Output fp32, Weight fp32, Input fp32
+#                  1: Output fp16, Weight fp16, Input fp16
+#                  2: Output bf16, Weight bf16, Input bf16
+# arg3: tensor layout (0: Output[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Input[G, N, Ho, Wo, K]
+#                      1: Output[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Input[N, Ho, Wo, G, K])
+# arg4: verification (0: no, 1: yes)
+# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
+# arg6: print tensor value (0: no; 1: yes)
+# arg7: time kernel (0: no, 1: yes)
+# Following arguments (depending on number of spatial dims):
+#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  G, N, K, C, 
+#  <filter spatial dimensions>, (ie Y, X for 2D)
+#  <input image spatial dimensions>, (ie Hi, Wi for 2D)
+#  <strides>, (ie Sy, Sx for 2D)
+#  <dilations>, (ie Dy, Dx for 2D)
+#  <left padding>, (ie LeftPy, LeftPx for 2D)
+#  <right padding>, (ie RightPy, RightPx for 2D)
+
+ ################                   op   datatype  layout  verify  init  log  time  Ndims  G  N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx
+./bin/ckProfiler grouped_conv_bwd_data          1       0       1     1    0     1      2 32  4 192 192  3  3  28  28   1   1   1   1       1       1        1        1
+
+ ```
+
+Result (MI100, FP16, GNHWC_GKYXC_GNHWK)
+```
+out: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
+wei: dim 5, lengths {32, 192, 192, 3, 3}, strides {331776, 1728, 1, 576, 192}
+in: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
+....
+Best configuration parameters:
+name: DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<256, 128, 256, 32, 8, 2, Default, 32, 32, 2, 4, 8, 4, 1, 1>
+avg_time: 0.768321
+tflops: 86.6679
+GB/s: 127.947
+```
+
+## Profile grouped convolution backward weight kernels
+```bash
+# arg1: tensor operation (grouped_conv_bwd_data: Grouped Convolution Backward Data)
+# arg2: data type (0: Input fp32, Weight fp32, Output fp32
+#                  1: Input fp16, Weight fp16, Output fp16
+#                  2: Input bf16, Weight fp32, Output bf16)
+# arg3: tensor layout (0: Input[G, N, C, Hi, Wi], Weight[G, K, C, Y, X], Output[G, N, K, Ho, Wo]
+#                      1: Input[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Output[G, N, Ho, Wo, K]
+#                      2: Input[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Output[N, Ho, Wo, G, K]
+# arg4: verification (0: no, 1: yes)
+# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
+# arg6: print tensor value (0: no; 1: yes)
+# arg7: time kernel (0: no, 1: yes)
+# Following arguments (depending on number of spatial dims):
+#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  G, N, K, C, 
+#  <filter spatial dimensions>, (ie Y, X for 2D)
+#  <input image spatial dimensions>, (ie Hi, Wi for 2D)
+#  <strides>, (ie Sy, Sx for 2D)
+#  <dilations>, (ie Dy, Dx for 2D)
+#  <left padding>, (ie LeftPy, LeftPx for 2D)
+#  <right padding>, (ie RightPy, RightPx for 2D)
+# SplitK
+
+ ################                   op   datatype  layout  verify  init  log  time  Ndims  G   N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx  SplitK
+./bin/ckProfiler grouped_conv_bwd_data          1       0       1     1    0     1      2 32 256 256 512  3  3  28  28   1   1   1   1       1       0        0        0       1
+
+ ```
+
+Result (MI100, FP16, GNHWC_GKYXC_GNHWK)
+```
+input: dim 5, lengths {32, 512, 1024, 28, 28}, strides {411041792, 802816, 1, 28672, 1024}
+weight: dim 5, lengths {32, 512, 1024, 3, 3}, strides {4718592, 9216, 1, 3072, 1024}
+output: dim 5, lengths {32, 512, 512, 26, 26}, strides {177209344, 346112, 1, 13312, 512}
+....
+Best configuration parameters:
+name: DeviceGroupedConvBwdWeight_Xdl_CShuffle<256, 256, 128, 4, Default, 8, 4, 2, 8, 4, 8, 2, 1, 1, 8>
+avg_time: 68.5216
+tflops: 95.337
+GB/s: 69.2301
+```
+Note: This kernel use atomic add, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
diff --git a/profiler/include/profiler/data_type_enum.hpp b/profiler/include/profiler/data_type_enum.hpp
index afcd6fea224f3e06f29309349bcc6ad55caa7d4f..c046c7fabb30e831fe41b31f936febef0f5bf148 100644
--- a/profiler/include/profiler/data_type_enum.hpp
+++ b/profiler/include/profiler/data_type_enum.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/data_type_enum_helper.hpp b/profiler/include/profiler/data_type_enum_helper.hpp
deleted file mode 100644
index d9bd5e1a4008fc5e6c09815c45408d8b07376a4f..0000000000000000000000000000000000000000
--- a/profiler/include/profiler/data_type_enum_helper.hpp
+++ /dev/null
@@ -1,77 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma
-
-#include "ck/utility/data_type.hpp"
-#include "profiler/data_type_enum.hpp"
-
-namespace ck {
-
-template <DataTypeEnum DataTypeEnum>
-struct get_datatype_from_enum;
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Int8>
-{
-    using type = int8_t;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Int32>
-{
-    using type = int32_t;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Half>
-{
-    using type = half_t;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Float>
-{
-    using type = float;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Double>
-{
-    using type = double;
-};
-
-template <typename T>
-struct get_datatype_enum_from_type;
-
-template <>
-struct get_datatype_enum_from_type<int8_t>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Int8;
-};
-
-template <>
-struct get_datatype_enum_from_type<int32_t>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Int32;
-};
-
-template <>
-struct get_datatype_enum_from_type<half_t>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Half;
-};
-
-template <>
-struct get_datatype_enum_from_type<float>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Float;
-};
-
-template <>
-struct get_datatype_enum_from_type<double>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Double;
-};
-
-} // namespace ck
diff --git a/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp b/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp
index b16254279ce433d06f71ca350b53841a294a3de6..22dab31100d469938d39de0db89bf5b2257383c8 100644
--- a/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp b/profiler/include/profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..5bee67c1ce9dc619f2b5208bf350a8fc01dceb06
--- /dev/null
+++ b/profiler/include/profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp
@@ -0,0 +1,395 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <index_t NumDimG,
+          index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          index_t NumDimO,
+          typename ADataType,
+          typename B0DataType,
+          typename B1DataType,
+          typename CDataType,
+          typename Acc0BiasesDataType,
+          typename Acc1BiasesDataType,
+          tensor_operation::device::MaskingSpecialization MaskingSpec>
+bool profile_batched_gemm_bias_softmax_gemm_permute_impl(bool do_verification,
+                                                         int init_method,
+                                                         bool do_log,
+                                                         bool time_kernel,
+                                                         int M,
+                                                         int N,
+                                                         int K,
+                                                         int O,
+                                                         int G0,
+                                                         int G1,
+                                                         float alpha = -1.f)
+
+{
+
+    using PassThrough   = tensor_operation::element_wise::PassThrough;
+    using ScaleAdd      = tensor_operation::element_wise::ScaleAdd;
+    using AElementOp    = PassThrough;
+    using B0ElementOp   = PassThrough;
+    using C0DEElementOp = ScaleAdd;
+    using Acc0ElementOp = PassThrough;
+    using B1ElementOp   = PassThrough;
+    using CElementOp    = PassThrough;
+    using AccDataType   = float;
+    using D0DataType    = tuple_element_t<0, Acc0BiasesDataType>;
+    using tensor_operation::device::MaskingSpecialization;
+
+    // Ref Gemm0: various type in, fp32 out
+    using ReferenceGemm0Instance = tensor_operation::host::ReferenceBatchedGemm<ADataType,
+                                                                                B0DataType,
+                                                                                AccDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                B0ElementOp,
+                                                                                Acc0ElementOp>;
+
+    // Ref Softmax: fp32 in, various type out
+    using ReferenceSoftmaxInstance =
+        tensor_operation::host::ReferenceSoftmax<AccDataType, ADataType, AccDataType>;
+
+    // Ref Gemm1: various type in, various type out
+    using ReferenceGemm1Instance = tensor_operation::host::ReferenceBatchedGemm<ADataType,
+                                                                                B1DataType,
+                                                                                CDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                B1ElementOp,
+                                                                                CElementOp>;
+
+    bool pass = true;
+
+    // A layout [G0, M, G1, K]
+    std::vector<ck::index_t> a_gs_ms_ks_lengths{G0, G1, M, K};
+    std::vector<ck::index_t> a_gs_ms_ks_strides{M * G1 * K, K, G1 * K, 1};
+
+    // B0 layout [G0, N, G1, K]
+    std::vector<ck::index_t> b0_gs_ns_ks_lengths{G0, G1, N, K};
+    std::vector<ck::index_t> b0_gs_ns_ks_strides{N * G1 * K, K, G1 * K, 1};
+
+    // B1 layout [G0, N, G1, O]
+    std::vector<ck::index_t> b1_gs_os_ns_lengths{G0, G1, O, N};
+    std::vector<ck::index_t> b1_gs_os_ns_strides{N * G1 * O, O, 1, G1 * O};
+
+    // C layout [G0, M, G1, O]
+    std::vector<ck::index_t> c_gs_ms_os_lengths{G0, G1, M, O};
+    std::vector<ck::index_t> c_gs_ms_os_strides{M * G1 * O, O, G1 * O, 1};
+
+    // D layout [G0, M, G1, N]
+    std::vector<ck::index_t> d0_gs_ms_ns_lengths{G0, G1, M, N};
+    std::vector<ck::index_t> d0_gs_ms_ns_strides{M * G1 * N, N, G1 * N, 1};
+
+    const int BatchCount = G0 * G1;
+
+    Tensor<ADataType> a_gs_ms_ks(a_gs_ms_ks_lengths, a_gs_ms_ks_strides);
+    Tensor<B0DataType> b0_gs_ns_ks(b0_gs_ns_ks_lengths, b0_gs_ns_ks_strides);
+    Tensor<D0DataType> d0_gs_ms_ns(d0_gs_ms_ns_lengths, d0_gs_ms_ns_strides);
+    Tensor<B1DataType> b1_gs_os_ns(b1_gs_os_ns_lengths, b1_gs_os_ns_strides);
+    Tensor<CDataType> c_gs_ms_os_host_result(c_gs_ms_os_lengths, c_gs_ms_os_strides);
+    Tensor<CDataType> c_gs_ms_os_device_result(c_gs_ms_os_lengths, c_gs_ms_os_strides);
+
+    std::cout << "a_gs_ms_ks: " << a_gs_ms_ks.mDesc << std::endl;
+    std::cout << "b0_gs_ns_ks: " << b0_gs_ns_ks.mDesc << std::endl;
+    std::cout << "b1_gs_os_ns: " << b1_gs_os_ns.mDesc << std::endl;
+    std::cout << "c_gs_ms_os: " << c_gs_ms_os_host_result.mDesc << std::endl;
+
+    std::srand(1); // work around test flakiness
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        // Still unsure whether this kind of deterministic floating point accurary issue is expected
+        // or not. May want to try exact same approach as the GPU kernel in the host reference
+        // GEMM+Softmax+GEMM function to see if the accuracy discrepancy goes away. Until then,
+        // shrink the input value range as it is less likely to produce errors of around ~1e-3.
+        // a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        // b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-5, 5});
+        // b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-5, 5});
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-2, 2});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_2<D0DataType>{-2, 2});
+        break;
+    case 2:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_3<B0DataType>{0.0, 1.0});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_3<B1DataType>{-0.5, 0.5});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_3<D0DataType>{-0.5, 0.5});
+        break;
+    case 3:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-2, 2});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Diagonal<B0DataType>{});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_1<D0DataType>{1});
+        break;
+    default:
+        a_gs_ms_ks.GenerateTensorValue(GeneratorTensor_1<ADataType>{1});
+        b0_gs_ns_ks.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+        b1_gs_os_ns.GenerateTensorValue(GeneratorTensor_Diagonal<B1DataType>{});
+        d0_gs_ms_ns.GenerateTensorValue(GeneratorTensor_1<D0DataType>{1});
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_gs_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b0_device_buf(sizeof(B0DataType) * b0_gs_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d0_device_buf(sizeof(D0DataType) * d0_gs_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem b1_device_buf(sizeof(B1DataType) * b1_gs_os_ns.mDesc.GetElementSpaceSize());
+    DeviceMem c_device_buf(sizeof(CDataType) *
+                           c_gs_ms_os_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_gs_ms_ks.mData.data());
+    b0_device_buf.ToDevice(b0_gs_ns_ks.mData.data());
+    b1_device_buf.ToDevice(b1_gs_os_ns.mData.data());
+    d0_device_buf.ToDevice(d0_gs_ms_ns.mData.data());
+
+    if(alpha < 0)
+    {
+        alpha = 1.f / std::sqrt(K); // usually 1 / sqrt(head_dim)
+    }
+    auto a_element_op    = AElementOp{};
+    auto b0_element_op   = B0ElementOp{};
+    auto c0de_element_op = C0DEElementOp{alpha};
+    auto acc0_element_op = Acc0ElementOp{};
+    auto b1_element_op   = B1ElementOp{};
+    auto c_element_op    = CElementOp{};
+
+    using DeviceOp =
+        tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute<2,
+                                                                      1,
+                                                                      1,
+                                                                      1,
+                                                                      1,
+                                                                      ADataType,
+                                                                      B0DataType,
+                                                                      B1DataType,
+                                                                      CDataType,
+                                                                      Acc0BiasesDataType,
+                                                                      ck::Tuple<>,
+                                                                      AElementOp,
+                                                                      B0ElementOp,
+                                                                      C0DEElementOp,
+                                                                      B1ElementOp,
+                                                                      CElementOp,
+                                                                      MaskingSpec>;
+
+    // get device op instances
+    const auto op_ptrs = tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    if(do_verification)
+    {
+        c_device_buf.FromDevice(c_gs_ms_os_device_result.mData.data());
+
+        Tensor<ADataType> a_g_m_k({BatchCount, M, K});
+        Tensor<B0DataType> b0_g_k_n({BatchCount, K, N});
+        Tensor<B1DataType> b1_g_n_o({BatchCount, N, O});
+        Tensor<AccDataType> acc0_g_m_n({BatchCount, M, N});        // scratch object after gemm0
+        Tensor<ADataType> a1_g_m_n({BatchCount, M, N});            // scratch object after softmax
+        Tensor<CDataType> c_g_m_o_host_result({BatchCount, M, O}); // scratch object after gemm1
+        Tensor<D0DataType> d0_g_m_n({BatchCount, M, N});
+
+        // permute
+        a_gs_ms_ks.ForEach([&](auto& self, auto idx) {
+            a_g_m_k(idx[0] * G1 + idx[1], idx[2], idx[3]) = self(idx);
+        });
+        b0_gs_ns_ks.ForEach([&](auto& self, auto idx) {
+            b0_g_k_n(idx[0] * G1 + idx[1], idx[3], idx[2]) = self(idx);
+        });
+        b1_gs_os_ns.ForEach([&](auto& self, auto idx) {
+            b1_g_n_o(idx[0] * G1 + idx[1], idx[3], idx[2]) = self(idx);
+        });
+        d0_gs_ms_ns.ForEach([&](auto& self, auto idx) {
+            d0_g_m_n(idx[0] * G1 + idx[1], idx[2], idx[3]) = self(idx);
+        });
+
+        auto ref_gemm0          = ReferenceGemm0Instance{};
+        auto ref_gemm0_invoker  = ref_gemm0.MakeInvoker();
+        auto ref_gemm0_argument = ref_gemm0.MakeArgument(
+            a_g_m_k, b0_g_k_n, acc0_g_m_n, a_element_op, b0_element_op, acc0_element_op);
+
+        ref_gemm0_invoker.Run(ref_gemm0_argument);
+
+        acc0_g_m_n.ForEach([&](auto&, auto idx) {
+            c0de_element_op(acc0_g_m_n(idx), acc0_g_m_n(idx), d0_g_m_n(idx));
+        });
+        // mask out upper triangle
+        acc0_g_m_n.ForEach([&](auto& self, auto idx) {
+            if(MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle && idx[1] < idx[2])
+                self(idx) = -ck::NumericLimits<float>::Infinity();
+        });
+
+        auto ref_softmax          = ReferenceSoftmaxInstance{};
+        auto ref_softmax_invoker  = ref_softmax.MakeInvoker();
+        auto ref_softmax_argument = ref_softmax.MakeArgument(acc0_g_m_n, a1_g_m_n, 1, 0, {2});
+
+        ref_softmax_invoker.Run(ref_softmax_argument);
+
+        auto ref_gemm1          = ReferenceGemm1Instance{};
+        auto ref_gemm1_invoker  = ref_gemm1.MakeInvoker();
+        auto ref_gemm1_argument = ref_gemm1.MakeArgument(
+            a1_g_m_n, b1_g_n_o, c_g_m_o_host_result, PassThrough{}, b1_element_op, c_element_op);
+
+        ref_gemm1_invoker.Run(ref_gemm1_argument);
+
+        // permute
+        c_gs_ms_os_host_result.ForEach([&](auto& self, auto idx) {
+            const size_t& g0 = idx[0];
+            const size_t& g1 = idx[1];
+
+            const size_t g = g0 * G1 + g1;
+
+            self(idx) = c_g_m_o_host_result(g, idx[2], idx[3]);
+        });
+    }
+
+    std::string best_op_name;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+            static_cast<B0DataType*>(b0_device_buf.GetDeviceBuffer()),
+            static_cast<B1DataType*>(b1_device_buf.GetDeviceBuffer()),
+            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+            std::array<void*, 1>{
+                d0_device_buf.GetDeviceBuffer()}, // std::array<void*, 1> p_acc0_biases;
+            {},                                   // std::array<void*, 1> p_acc1_biases;
+            a_gs_ms_ks_lengths,
+            a_gs_ms_ks_strides,
+            b0_gs_ns_ks_lengths,
+            b0_gs_ns_ks_strides,
+            b1_gs_os_ns_lengths,
+            b1_gs_os_ns_strides,
+            c_gs_ms_os_lengths,
+            c_gs_ms_os_strides,
+            std::array<std::vector<ck::index_t>, 1>{
+                d0_gs_ms_ns_lengths}, // acc0_biases_gs_ms_ns_lengths
+            std::array<std::vector<ck::index_t>, 1>{
+                d0_gs_ms_ns_strides}, // std::array<std::vector<ck::index_t>,
+                                      // 1>{acc0_biases_gs_ms_ns_strides},
+            {}, // std::array<std::vector<ck::index_t>, 1>{acc1_biases_gs_ms_os_lengths},
+            {}, // std::array<std::vector<ck::index_t>, 1>{acc1_biases_gs_ms_os_strides},
+            a_element_op,
+            b0_element_op,
+            c0de_element_op,
+            b1_element_op,
+            c_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            std::string op_name = op_ptr->GetTypeString();
+
+            float ave_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop      = (size_t(M) * N * K * 2 + size_t(M) * N * O * 2) * BatchCount;
+            std::size_t num_btype = (sizeof(ADataType) * M * K + sizeof(B0DataType) * K * N +
+                                     sizeof(B1DataType) * N * O + sizeof(CDataType) * M * O +
+                                     sizeof(D0DataType) * M * N) *
+                                    BatchCount;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                      << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                c_device_buf.FromDevice(c_gs_ms_os_device_result.mData.data());
+
+                // default absolute error and relative error is 0.001
+                double rtol = 1e-3;
+                double atol = 1e-3;
+
+                // when BF16 is taken, set absolute error and relative error to 0.01
+                if(std::is_same_v<ADataType, ck::bhalf_t> &&
+                   std::is_same_v<B0DataType, ck::bhalf_t> &&
+                   std::is_same_v<B1DataType, ck::bhalf_t> &&
+                   std::is_same_v<CDataType, ck::bhalf_t> &&
+                   std::is_same_v<D0DataType, ck::bhalf_t>)
+                {
+                    rtol = 1e-2;
+                    atol = 1e-2;
+                }
+
+                pass = pass & ck::utils::check_err(c_gs_ms_os_device_result,
+                                                   c_gs_ms_os_host_result,
+                                                   "Error: Incorrect results!",
+                                                   rtol,
+                                                   atol);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "a_gs_ms_ks: ", a_gs_ms_ks.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "b0_gs_ns_ks : ", b0_gs_ns_ks.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "b1_gs_os_ns : ", b1_gs_os_ns.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(
+                        std::cout << "c_gs_ms_os_host_result : ", c_gs_ms_os_host_result.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_gs_ms_os_device_result : ",
+                                          c_gs_ms_os_device_result.mData,
+                                          ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp b/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp
index 1583c6db21e0959bd155946b2e81363091b4f43f..f3d2c5561756d43a4af44bbde12cf93c26f5602c 100644
--- a/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_impl.hpp b/profiler/include/profiler/profile_batched_gemm_impl.hpp
index c07d7c0555490559bbf0357a5d2657bcbe5261fd..936c22f5d89143198a1d4a01cb06bd5c66ca043e 100644
--- a/profiler/include/profiler/profile_batched_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -8,9 +8,11 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
@@ -27,7 +29,11 @@ template <typename ADataType,
           typename CDataType,
           typename ALayout,
           typename BLayout,
-          typename CLayout>
+          typename CLayout,
+          typename AElementOp,
+          typename BElementOp,
+          typename CElementOp,
+          typename DeviceOp>
 bool profile_batched_gemm_impl(int do_verification,
                                int init_method,
                                bool do_log,
@@ -88,10 +94,6 @@ bool profile_batched_gemm_impl(int do_verification,
         b_g_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
     }
 
-    using AElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using BElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using CElementOp = ck::tensor_operation::element_wise::PassThrough;
-
     const auto a_element_op = AElementOp{};
     const auto b_element_op = BElementOp{};
     const auto c_element_op = CElementOp{};
@@ -124,16 +126,6 @@ bool profile_batched_gemm_impl(int do_verification,
     b_device_buf.ToDevice(b_g_k_n.mData.data());
     c_device_buf.ToDevice(c_g_m_n_device_result.mData.data());
 
-    using DeviceOp = ck::tensor_operation::device::DeviceBatchedGemm<ALayout,
-                                                                     BLayout,
-                                                                     CLayout,
-                                                                     ADataType,
-                                                                     BDataType,
-                                                                     CDataType,
-                                                                     AElementOp,
-                                                                     BElementOp,
-                                                                     CElementOp>;
-
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
         DeviceOp>::GetInstances();
@@ -148,23 +140,62 @@ bool profile_batched_gemm_impl(int do_verification,
     // profile device op instances
     for(auto& op_ptr : op_ptrs)
     {
-        auto argument_ptr =
-            op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
-                                        static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
-                                        static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
-                                        M,
-                                        N,
-                                        K,
-                                        StrideA,
-                                        StrideB,
-                                        StrideC,
-                                        BatchStrideA,
-                                        BatchStrideB,
-                                        BatchStrideC,
-                                        BatchCount,
-                                        ck::tensor_operation::element_wise::PassThrough{},
-                                        ck::tensor_operation::element_wise::PassThrough{},
-                                        ck::tensor_operation::element_wise::PassThrough{});
+        std::unique_ptr<tensor_operation::device::BaseArgument> argument_ptr;
+        // false branch for multi d dl kernel
+        if constexpr(std::is_same<
+                         DeviceOp,
+                         ck::tensor_operation::device::DeviceBatchedGemm<ALayout,
+                                                                         BLayout,
+                                                                         CLayout,
+                                                                         ADataType,
+                                                                         BDataType,
+                                                                         CDataType,
+                                                                         AElementOp,
+                                                                         BElementOp,
+                                                                         CElementOp>>::value)
+        {
+
+            argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+                                            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+                                            M,
+                                            N,
+                                            K,
+                                            StrideA,
+                                            StrideB,
+                                            StrideC,
+                                            BatchStrideA,
+                                            BatchStrideB,
+                                            BatchStrideC,
+                                            BatchCount,
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{});
+        }
+        else
+        {
+            argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+                                            {},
+                                            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+                                            M,
+                                            N,
+                                            K,
+                                            BatchCount,
+                                            StrideA,
+                                            StrideB,
+                                            {},
+                                            StrideC,
+                                            BatchStrideA,
+                                            BatchStrideB,
+                                            {},
+                                            BatchStrideC,
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{});
+        }
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
diff --git a/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp b/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp
index 45b7b77388b63c8abba43aabc50a9c0dd86eeede..901fa338d4d7aab26a70da548e2b000c98840d26 100644
--- a/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp
index f5ec235141a7238157c0d2fd34f1f9c849decfc4..15a21206c5945d700069cd7a8f8a9fdff7c4e9bf 100644
--- a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp
index 91c28f25fc5953460a00d87961b97d1587aad1ce..f2fcb0b133861b87a92e3434c8c3e22b0ef1a51c 100644
--- a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batchnorm_backward_impl.hpp b/profiler/include/profiler/profile_batchnorm_backward_impl.hpp
index 79d8862081fd78b0706110219dc5af5016cd9122..3343b5e66e4b8666edf778abf4393dd75cbd0937 100644
--- a/profiler/include/profiler/profile_batchnorm_backward_impl.hpp
+++ b/profiler/include/profiler/profile_batchnorm_backward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batchnorm_forward_impl.hpp b/profiler/include/profiler/profile_batchnorm_forward_impl.hpp
index 82fe75bf015beba4d56898f67aa2100b2d3c6ff4..2f9538b16c326a740b1e3bd0c3c150eaaadb9ebf 100644
--- a/profiler/include/profiler/profile_batchnorm_forward_impl.hpp
+++ b/profiler/include/profiler/profile_batchnorm_forward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batchnorm_infer_impl.hpp b/profiler/include/profiler/profile_batchnorm_infer_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..1b31a2aabf5f42a115d83270f91d87f0f74d8a4d
--- /dev/null
+++ b/profiler/include/profiler/profile_batchnorm_infer_impl.hpp
@@ -0,0 +1,335 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+#include <stdexcept>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename XDataType,
+          typename YDataType,
+          typename AccDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          index_t Rank,
+          index_t NumBatchNormReduceDim>
+bool profile_batchnorm_infer_impl(int do_verification,
+                                  int init_method,
+                                  bool do_dumpout,
+                                  bool time_kernel,
+                                  const std::vector<size_t> inOutLengths,
+                                  const std::vector<int> reduceDims,
+                                  double epsilon)
+{
+    if(inOutLengths.size() != Rank || reduceDims.size() != NumBatchNormReduceDim)
+    {
+        throw std::runtime_error("Invalid tensor lengths or number of reduce dimensions!");
+    };
+
+    std::vector<size_t> scaleBiasMeanVarLengths;
+    std::vector<int> invariantDims;
+
+    // used for calculating the effective transferred bytes by each operation
+    size_t total_length;
+    size_t invariant_length = 1;
+
+    total_length =
+        std::accumulate(inOutLengths.begin(), inOutLengths.end(), 1, std::multiplies<size_t>{});
+
+    if(std::any_of(reduceDims.begin(), reduceDims.end(), [](int d) { return d < 0 || d >= Rank; }))
+        throw std::runtime_error("Invalid reduce dimensions!");
+
+    for(int dim = 0; dim < Rank; dim++)
+    {
+        if(std::none_of(reduceDims.begin(), reduceDims.end(), [&](int d) { return dim == d; }))
+        {
+            invariantDims.push_back(dim);
+            scaleBiasMeanVarLengths.push_back(inOutLengths[dim]);
+            invariant_length *= inOutLengths[dim];
+        };
+    }
+
+    // input data of the batchnorm infer algorithm
+    Tensor<XDataType> x(inOutLengths);
+    Tensor<ScaleDataType> scale(scaleBiasMeanVarLengths);
+    Tensor<BiasDataType> bias(scaleBiasMeanVarLengths);
+    Tensor<MeanVarDataType> estimatedMean(scaleBiasMeanVarLengths);
+    Tensor<MeanVarDataType> estimatedVariance(scaleBiasMeanVarLengths);
+
+    // output data of the batchnorm infer algorithm
+    Tensor<YDataType> y_ref(inOutLengths);
+    Tensor<YDataType> y(inOutLengths);
+
+    auto inOutStrides            = x.mDesc.GetStrides();
+    auto scaleBiasMeanVarStrides = scale.mDesc.GetStrides();
+
+    std::size_t num_thread = std::thread::hardware_concurrency();
+
+    const float x_mean       = 0.0f;
+    const float x_stddev     = 1.0f;
+    const float noise_stddev = 0.04f;
+
+    // input data in normal distribution
+    x.GenerateTensorValue(GeneratorTensor_4<XDataType>{x_mean, x_stddev}, num_thread);
+
+    // initialize the estimatedMean to be values with tiny variation to the mean of the x
+    // values
+    estimatedMean.GenerateTensorValue(GeneratorTensor_4<MeanVarDataType>{x_mean, noise_stddev},
+                                      num_thread);
+
+    // initialize the estimatedVariance to be values with tiny variation to the variance of
+    // the x values
+    estimatedVariance.GenerateTensorValue(
+        GeneratorTensor_4<MeanVarDataType>{x_stddev * x_stddev, noise_stddev}, num_thread);
+
+    if(do_verification)
+    {
+        switch(init_method)
+        {
+        case 0:
+            scale.GenerateTensorValue(GeneratorTensor_0<ScaleDataType>{}, num_thread);
+            bias.GenerateTensorValue(GeneratorTensor_0<BiasDataType>{}, num_thread);
+            break;
+        case 1:
+            scale.GenerateTensorValue(GeneratorTensor_1<ScaleDataType>{1}, num_thread);
+            bias.GenerateTensorValue(GeneratorTensor_1<BiasDataType>{0}, num_thread);
+            break;
+        case 2:
+            scale.GenerateTensorValue(GeneratorTensor_2<ScaleDataType>{-5, 5}, num_thread);
+            bias.GenerateTensorValue(GeneratorTensor_2<BiasDataType>{-5, 5}, num_thread);
+            break;
+        default:
+            scale.GenerateTensorValue(GeneratorTensor_3<ScaleDataType>{-1.0f, 1.0f}, num_thread);
+            bias.GenerateTensorValue(GeneratorTensor_3<BiasDataType>{-1.0f, 1.0f}, num_thread);
+        }
+    };
+
+    // these buffers are usually provided by the user application
+    DeviceMem x_dev(sizeof(XDataType) * x.mDesc.GetElementSpaceSize());
+    DeviceMem y_dev(sizeof(XDataType) * y.mDesc.GetElementSpaceSize());
+    DeviceMem scale_dev(sizeof(ScaleDataType) * scale.mDesc.GetElementSpaceSize());
+    DeviceMem bias_dev(sizeof(BiasDataType) * bias.mDesc.GetElementSpaceSize());
+
+    // estimatedMean_dev
+    DeviceMem estimatedMean_dev(sizeof(MeanVarDataType) *
+                                estimatedMean.mDesc.GetElementSpaceSize());
+    // estimatedVariance_dev
+    DeviceMem estimatedVariance_dev(sizeof(MeanVarDataType) *
+                                    estimatedVariance.mDesc.GetElementSpaceSize());
+
+    x_dev.ToDevice(x.mData.data());
+    scale_dev.ToDevice(scale.mData.data());
+    bias_dev.ToDevice(bias.mData.data());
+    estimatedMean_dev.ToDevice(estimatedMean.mData.data());
+    estimatedVariance_dev.ToDevice(estimatedVariance.mData.data());
+
+    std::array<index_t, Rank> arrInOutLengths;
+    std::array<index_t, Rank> arrInOutStrides;
+    std::array<index_t, Rank - NumBatchNormReduceDim> arrScaleBiasMeanVarLengths;
+    std::array<index_t, Rank - NumBatchNormReduceDim> arrScaleBiasMeanVarStrides;
+    std::array<int, NumBatchNormReduceDim> arrReduceDims;
+
+    std::copy(inOutLengths.begin(), inOutLengths.end(), arrInOutLengths.begin());
+    std::copy(inOutStrides.begin(), inOutStrides.end(), arrInOutStrides.begin());
+    std::copy(scaleBiasMeanVarLengths.begin(),
+              scaleBiasMeanVarLengths.end(),
+              arrScaleBiasMeanVarLengths.begin());
+    std::copy(scaleBiasMeanVarStrides.begin(),
+              scaleBiasMeanVarStrides.end(),
+              arrScaleBiasMeanVarStrides.begin());
+
+    std::copy(reduceDims.begin(), reduceDims.end(), arrReduceDims.begin());
+
+    std::array<index_t, Rank> aligned_scaleBiasMeanVarStrides{0};
+
+    int i = 0;
+    for(auto dim : invariantDims)
+    {
+        assert(inOutLengths[dim] == scaleBiasMeanVarLengths[i]);
+
+        aligned_scaleBiasMeanVarStrides[dim] = scaleBiasMeanVarStrides[i];
+        i++;
+    };
+
+    using Normalize = ck::tensor_operation::element_wise::NormalizeInInfer;
+
+    // add device batchnorm-infer instances
+    using DeviceOp = ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<XDataType, MeanVarDataType, MeanVarDataType, ScaleDataType, BiasDataType>,
+        ck::Tuple<YDataType>,
+        Normalize,
+        Rank>;
+
+    // get device op instances
+    const auto instance_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
+
+    std::string best_instance_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    if(do_verification)
+    {
+        using PassThroughOp = ck::tensor_operation::element_wise::PassThrough;
+
+        using ReferenceBatchNormInferInstance =
+            ck::tensor_operation::host::ReferenceBatchNormInfer<XDataType,
+                                                                YDataType,
+                                                                AccDataType,
+                                                                ScaleDataType,
+                                                                BiasDataType,
+                                                                MeanVarDataType,
+                                                                PassThroughOp,
+                                                                Rank,
+                                                                NumBatchNormReduceDim>;
+        auto batchNormInfer_ref = ReferenceBatchNormInferInstance{};
+
+        auto argument_ptr_ref =
+            batchNormInfer_ref.MakeArgumentPointer(arrInOutLengths,
+                                                   arrInOutStrides,
+                                                   arrInOutStrides,
+                                                   arrReduceDims,
+                                                   arrScaleBiasMeanVarLengths,
+                                                   arrScaleBiasMeanVarStrides,
+                                                   arrScaleBiasMeanVarStrides,
+                                                   arrScaleBiasMeanVarStrides,
+                                                   x.mData.data(),
+                                                   scale.mData.data(),
+                                                   bias.mData.data(),
+                                                   epsilon,
+                                                   PassThroughOp{},
+                                                   estimatedMean.mData.data(),
+                                                   estimatedVariance.mData.data(),
+                                                   y_ref.mData.data());
+
+        if(!batchNormInfer_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters not supported by the reference instance, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = batchNormInfer_ref.MakeInvokerPointer();
+
+        (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
+    }
+
+    int num_kernel = 0;
+    bool pass      = true;
+
+    for(auto& inst_ptr : instance_ptrs)
+    {
+        auto argument_ptr = inst_ptr->MakeArgumentPointer(arrInOutLengths,
+                                                          {arrInOutStrides,
+                                                           aligned_scaleBiasMeanVarStrides,
+                                                           aligned_scaleBiasMeanVarStrides,
+                                                           aligned_scaleBiasMeanVarStrides,
+                                                           aligned_scaleBiasMeanVarStrides},
+                                                          {arrInOutStrides},
+                                                          {x_dev.GetDeviceBuffer(),
+                                                           estimatedMean_dev.GetDeviceBuffer(),
+                                                           estimatedVariance_dev.GetDeviceBuffer(),
+                                                           scale_dev.GetDeviceBuffer(),
+                                                           bias_dev.GetDeviceBuffer()},
+                                                          {y_dev.GetDeviceBuffer()},
+                                                          Normalize{epsilon});
+
+        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            num_kernel++;
+        }
+        else
+        {
+            if(time_kernel)
+            {
+                std::cout << inst_ptr->GetTypeString()
+                          << " skipped due to unsupported argument: " << std::endl;
+            }
+
+            continue;
+        };
+
+        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
+
+        float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        size_t num_bytes = 0;
+
+        // inputing of x, scale, bias, outputing of y
+        num_bytes += total_length * (sizeof(XDataType) + sizeof(YDataType)) +
+                     invariant_length *
+                         (sizeof(ScaleDataType) + sizeof(BiasDataType) + sizeof(MeanVarDataType));
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        if(time_kernel)
+            std::cout << "Perf: " << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << inst_ptr->GetTypeString() << std::endl;
+
+        if(avg_time < best_avg_time)
+        {
+            best_instance_name = inst_ptr->GetTypeString();
+            best_avg_time      = avg_time;
+            best_gb_per_sec    = gb_per_sec;
+        }
+
+        if(do_verification)
+        {
+            using ck::utils::check_err;
+            bool single_pass;
+
+            y_dev.FromDevice(y.mData.data());
+
+            if constexpr(ck::is_same_v<YDataType, ck::bhalf_t>)
+                single_pass = check_err(y.mData, y_ref.mData, "y results", 1e-2, 1e-2);
+            else
+                single_pass = check_err(y.mData, y_ref.mData, "y results", 4e-3, 4e-3);
+
+            pass = pass && single_pass;
+        };
+
+        if(do_dumpout)
+        {
+            using ck::host_common::dumpBufferToFile;
+
+            // clang-format off
+            dumpBufferToFile("dump_x.bin", x.mData.data(), x.mDesc.GetElementSize());
+            dumpBufferToFile("dump_y.bin", y.mData.data(), y.mDesc.GetElementSize());
+            dumpBufferToFile("dump_y_ref.bin", y_ref.mData.data(), y_ref.mDesc.GetElementSize());
+            // clang-format off
+        };
+    }
+
+    if(time_kernel)
+    {
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
+    }
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_contraction_impl.hpp b/profiler/include/profiler/profile_contraction_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..660cc3f9e5236fdba5ce3c6373189cbc1ed5fbc1
--- /dev/null
+++ b/profiler/include/profiler/profile_contraction_impl.hpp
@@ -0,0 +1,345 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+#include <iostream>
+#include <typeinfo>
+#include <limits>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction_scale.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+#include "ck/host_utility/io.hpp"
+
+namespace ck {
+namespace profiler {
+
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+using Scale    = ck::tensor_operation::element_wise::Scale;
+
+template <typename ALayout,
+          typename BLayout,
+          typename CDELayout,
+          typename DataType,
+          typename DTupleDataType,
+          typename CDElementOp>
+int profile_contraction_impl(ck::index_t do_verification,
+                             ck::index_t init_method,
+                             bool do_log,
+                             bool time_kernel,
+                             CDElementOp cde_element_op,
+                             const std::vector<ck::index_t>& M,
+                             const std::vector<ck::index_t>& N,
+                             const std::vector<ck::index_t>& K,
+                             const std::vector<ck::index_t>& StridesA,
+                             const std::vector<ck::index_t>& StridesB,
+                             const std::vector<ck::index_t>& StridesE,
+                             const std::vector<ck::index_t>& StridesD)
+{
+    bool pass = true;
+
+    auto f_host_tensor_descriptor = [](const std::vector<ck::index_t>& dims01,
+                                       const std::vector<ck::index_t>& dims23,
+                                       const std::vector<ck::index_t>& strides) {
+        std::vector<std::size_t> dims_szt(dims01.begin(), dims01.end());
+        dims_szt.insert(dims_szt.end(), dims23.begin(), dims23.end());
+        std::vector<std::size_t> strides_szt(strides.begin(), strides.end());
+
+        return HostTensorDescriptor(dims_szt, strides);
+    };
+
+    Tensor<DataType> a_m_k(f_host_tensor_descriptor(M, K, StridesA));
+    Tensor<DataType> b_k_n(f_host_tensor_descriptor(K, N, StridesB));
+    Tensor<DataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
+    Tensor<DataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StridesE));
+    Tensor<DataType> d_m_n(f_host_tensor_descriptor(M, N, StridesD));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_device_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<DataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
+    }
+
+    using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+    DeviceMem a_device_buf(sizeof(DataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(DataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(DataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DataType) * d_m_n.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    e_device_buf.SetZero();
+    d_device_buf.ToDevice(d_m_n.mData.data());
+
+    const std::vector<index_t> a_ms_ks_lengths = {M[0], M[1], K[0], K[1]};
+    const std::vector<index_t> b_ns_ks_lengths = {N[0], N[1], K[0], K[1]};
+    const std::vector<index_t> e_ms_ns_lengths = {M[0], M[1], N[0], N[1]};
+    const std::vector<index_t> d_m_n_lengths   = {M[0], M[1], N[0], N[1]};
+
+    const auto a_element_op = AElementOp{};
+    const auto b_element_op = BElementOp{};
+
+    constexpr ck::index_t NumDim = 2;
+    using DeviceOp               = ck::tensor_operation::device::DeviceContractionMultipleD<NumDim,
+                                                                              NumDim,
+                                                                              NumDim,
+                                                                              DataType,
+                                                                              DataType,
+                                                                              DTupleDataType,
+                                                                              DataType,
+                                                                              AElementOp,
+                                                                              BElementOp,
+                                                                              CDElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    // Run reference op
+    if(do_verification)
+    {
+        using ReferenceGemmInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDim,
+                                                                      NumDim,
+                                                                      NumDim,
+                                                                      DataType,
+                                                                      DataType,
+                                                                      DataType,
+                                                                      DataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceGemmInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<DataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_m_n_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_m_n_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_m_n_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_m_n_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        if constexpr(is_same<CDElementOp, Bilinear>::value)
+                        {
+                            cde_element_op(e_m_n_host_result(m0, m1, n0, n1),
+                                           c_m_n_host_result(m0, m1, n0, n1),
+                                           d_m_n(m0, m1, n0, n1));
+                        }
+                        else if constexpr(is_same<CDElementOp, Scale>::value)
+                        {
+                            cde_element_op(e_m_n_host_result(m0, m1, n0, n1),
+                                           c_m_n_host_result(m0, m1, n0, n1));
+                        }
+                        else
+                        {
+                            static_assert("Unsupported CDElementOp in contraction profiler.");
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    std::string best_op_name;
+    float best_avg_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        std::unique_ptr<tensor_operation::device::BaseArgument> argument_ptr;
+        if constexpr(is_same<CDElementOp, Bilinear>::value)
+        {
+            argument_ptr = op_ptr->MakeArgumentPointer(
+                static_cast<DataType*>(a_device_buf.GetDeviceBuffer()),
+                static_cast<DataType*>(b_device_buf.GetDeviceBuffer()),
+                std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                static_cast<DataType*>(e_device_buf.GetDeviceBuffer()),
+                a_ms_ks_lengths,
+                StridesA,
+                b_ns_ks_lengths,
+                StridesB,
+                std::array<std::vector<ck::index_t>, 1>{d_m_n_lengths},
+                std::array<std::vector<ck::index_t>, 1>{StridesD},
+                e_ms_ns_lengths,
+                StridesE,
+                a_element_op,
+                b_element_op,
+                cde_element_op);
+        }
+        else if constexpr(is_same<CDElementOp, Scale>::value)
+        {
+            argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<DataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<DataType*>(b_device_buf.GetDeviceBuffer()),
+                                            std::array<const void*, 0>{},
+                                            static_cast<DataType*>(e_device_buf.GetDeviceBuffer()),
+                                            a_ms_ks_lengths,
+                                            StridesA,
+                                            b_ns_ks_lengths,
+                                            StridesB,
+                                            std::array<std::vector<ck::index_t>, 0>{},
+                                            std::array<std::vector<ck::index_t>, 0>{},
+                                            e_ms_ns_lengths,
+                                            StridesE,
+                                            a_element_op,
+                                            b_element_op,
+                                            cde_element_op);
+        }
+        else
+        {
+            static_assert("Unsupported CDElementOp in contraction profiler.");
+        }
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        auto nelems_m = M[0] * M[1];
+        auto nelems_n = N[0] * N[1];
+        auto nelems_k = K[0] * K[1];
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init C to zero before profiling next kernel
+            e_device_buf.SetZero();
+
+            std::string op_name = op_ptr->GetTypeString();
+
+            float avg_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop = std::size_t(2) * nelems_m * nelems_n * nelems_k;
+
+            std::size_t num_btype = sizeof(DataType) * nelems_m * nelems_k +
+                                    sizeof(DataType) * nelems_k * nelems_n +
+                                    sizeof(DataType) * nelems_m * nelems_n;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
+
+            float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+                float threshold =
+                    static_cast<DataType>(nelems_k) * std::numeric_limits<DataType>::epsilon();
+                pass = pass & ck::utils::check_err(e_m_n_device_result,
+                                                   e_m_n_host_result,
+                                                   "Error: incorrect results!",
+                                                   threshold,
+                                                   threshold);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_host  : ", e_m_n_host_result.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_device: ", e_m_n_device_result.mData, ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    }
+
+    if constexpr(is_same<DataType, float>::value)
+    {
+        std::cout << "Best Perf for datatype = f32";
+    }
+    else if constexpr(is_same<DataType, double>::value)
+    {
+        std::cout << "Best Perf for datatype = f64";
+    }
+
+    if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " ALayout =  RowMajor";
+    }
+    else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " ALayout =  ColumnMajor";
+    }
+
+    if constexpr(is_same<BLayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " BLayout =  RowMajor";
+    }
+    else if constexpr(is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " BLayout =  ColumnMajor";
+    }
+
+    if constexpr(is_same<CDELayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " CDELayout =  RowMajor";
+    }
+    else if constexpr(is_same<CDELayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " CDELayout =  ColumnMajor";
+    }
+
+    std::cout << " M = " << M << " N = " << N << " K = " << K << " StridesA = " << StridesA
+              << " StridesB = " << StridesB << " StridesE = " << StridesE << " : " << best_avg_time
+              << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec << " GB/s, "
+              << best_op_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_contraction_utils.hpp b/profiler/include/profiler/profile_contraction_utils.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..076bbd4559f6ec1652559030fa81be416385e962
--- /dev/null
+++ b/profiler/include/profiler/profile_contraction_utils.hpp
@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/ck.hpp"
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+using Scale    = ck::tensor_operation::element_wise::Scale;
+
+enum struct ContractionMatrixLayout
+{
+    MK_KN_MN_MN, // 0
+    MK_NK_MN_MN, // 1
+    KM_KN_MN_MN, // 2
+    KM_NK_MN_MN, // 3
+};
+
+enum struct ContractionDataType
+{
+    F32_F32_F32_F32, // 0
+    F64_F64_F64_F64, // 1
+};
+
+inline void collect_index_params(char* argv[],
+                                 std::vector<ck::index_t>& params,
+                                 const ck::index_t from,
+                                 const ck::index_t num)
+{
+    for(ck::index_t p = from; p < from + num; p++)
+        params.push_back(std::stoi(argv[p]));
+}
+
+// Defualt strides for row-major: {Dim1 * Dim2 * Dim3, Dim2 * Dim3, Dim3, 1}
+// Defualt strides for column-major: {Dim1, 1, Dim0 * Dim1 * Dim3, Dim0 * Dim1}
+inline void
+assign_default_strides(Row, std::vector<ck::index_t>& strides, std::vector<ck::index_t> dims)
+{
+    strides = {dims[1] * dims[2] * dims[3], dims[2] * dims[3], dims[3], 1};
+}
+
+inline void
+assign_default_strides(Col, std::vector<ck::index_t>& strides, std::vector<ck::index_t> dims)
+{
+    strides = {dims[1], 1, dims[0] * dims[1] * dims[3], dims[0] * dims[1]};
+}
diff --git a/profiler/include/profiler/profile_conv_bwd_data_impl.hpp b/profiler/include/profiler/profile_conv_bwd_data_impl.hpp
index 86d394daf90c79d6f9faf34254901dc8e2bd43e2..52152a90fe5eab2fd80ab156d9080c75b65d9eb0 100644
--- a/profiler/include/profiler/profile_conv_bwd_data_impl.hpp
+++ b/profiler/include/profiler/profile_conv_bwd_data_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp b/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp
index 1aebef8bb2b2b34a640797ab9e5985d5cb09fbea..436fbdbd759195add8fba952235392d31b9613a0 100644
--- a/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp
+++ b/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp b/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp
index 2bac144334eefc2d39ff63e17637b50d977921ec..808c1a1c901293816bf3fb153c8f8a7d2faaaf15 100644
--- a/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp
+++ b/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_conv_fwd_impl.hpp b/profiler/include/profiler/profile_conv_fwd_impl.hpp
index 1f3ba8f00714449e51f180879f3b38d01221cefe..bc2eb257970d1b91ab7fbf1a592d18df0ef3d159 100644
--- a/profiler/include/profiler/profile_conv_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_conv_fwd_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_convnd_bwd_data_impl.hpp b/profiler/include/profiler/profile_convnd_bwd_data_impl.hpp
deleted file mode 100644
index 1e69ebc8bd173b25311b45305a24d71fcd7324a9..0000000000000000000000000000000000000000
--- a/profiler/include/profiler/profile_convnd_bwd_data_impl.hpp
+++ /dev/null
@@ -1,486 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/conv_util.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
-
-using F16  = ck::half_t;
-using F32  = float;
-using BF16 = ck::bhalf_t;
-using INT8 = int8_t;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using DeviceConvBwdDataNoOpPtr =
-    DeviceConvBwdDataPtr<ck::tensor_operation::element_wise::PassThrough,
-                         ck::tensor_operation::element_wise::PassThrough,
-                         ck::tensor_operation::element_wise::PassThrough>;
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-namespace ck {
-namespace profiler {
-using DeviceConvBwdDataNoOpPtr = ck::tensor_operation::device::instance::DeviceConvBwdDataNoOpPtr;
-
-template <typename InLayout>
-HostTensorDescriptor get_input_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-template <typename WeiLayout>
-HostTensorDescriptor get_filters_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                        int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-template <typename OutLayout>
-HostTensorDescriptor get_output_host_ensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-template <typename InDataType, typename WeiDataType, typename OutDataType>
-void get_device_conv_bwd_data_op_ptr(
-    InDataType, WeiDataType, OutDataType, std::vector<DeviceConvBwdDataNoOpPtr>&, int)
-{
-    std::cout << "can not find device conv bwd data" << std::endl;
-    exit(1);
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    F32, F32, F32, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    F16, F16, F16, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    BF16, BF16, BF16, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    INT8, INT8, INT8, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <typename T>
-static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
-{
-    float max_diff = 1e-6;
-
-    for(std::size_t i = 0; i < ref.mData.size(); ++i)
-    {
-        float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
-        if(max_diff < diff)
-        {
-            return false;
-        }
-    }
-    return true;
-}
-template <typename DataType>
-void show_data_nhwc_layout(Tensor<DataType>& nhwc)
-{
-    std::cout << "[";
-    for(int n = 0; n < ck::type_convert<int>(nhwc.mDesc.GetLengths()[0]); n++)
-    {
-        std::cout << "[";
-        for(int hi = 0; hi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[2]); hi++)
-        {
-            std::cout << "[";
-            for(int wi = 0; wi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[3]); wi++)
-            {
-                std::cout << "[";
-                for(int c = 0; c < ck::type_convert<int>(nhwc.mDesc.GetLengths()[1]); c++)
-                {
-                    std::cout << static_cast<float>(nhwc(n, c, hi, wi)) << "  ";
-                }
-                std::cout << "]";
-            }
-            std::cout << "]";
-        }
-        std::cout << "]";
-    }
-    std::cout << "]";
-}
-
-template <int NDimSpatial,
-          typename InDataType,
-          typename WeiDataType,
-          typename OutDataType,
-          typename AccDataType,
-          typename InLayout,
-          typename WeiLayout,
-          typename OutLayout>
-bool profile_convnd_bwd_data_impl(int do_verification,
-                                  int init_method,
-                                  bool do_log,
-                                  bool time_kernel,
-                                  ck::index_t N,
-                                  ck::index_t K,
-                                  ck::index_t C,
-                                  const std::vector<ck::index_t>& input_spatial_lengths,
-                                  const std::vector<ck::index_t>& filter_spatial_lengths,
-                                  const std::vector<ck::index_t>& output_spatial_lengths,
-                                  const std::vector<ck::index_t>& conv_filter_strides,
-                                  const std::vector<ck::index_t>& conv_filter_dilations,
-                                  const std::vector<ck::index_t>& input_left_pads,
-                                  const std::vector<ck::index_t>& input_right_pads)
-{
-    using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-    using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-    const auto in_element_op  = InElementOp{};
-    const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{};
-
-    std::vector<std::size_t> input_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(C)};
-    input_dims.insert(
-        std::end(input_dims), std::begin(input_spatial_lengths), std::end(input_spatial_lengths));
-
-    std::vector<std::size_t> filter_dims{static_cast<std::size_t>(K), static_cast<std::size_t>(C)};
-    filter_dims.insert(std::end(filter_dims),
-                       std::begin(filter_spatial_lengths),
-                       std::end(filter_spatial_lengths));
-
-    std::vector<std::size_t> output_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(K)};
-    output_dims.insert(std::end(output_dims),
-                       std::begin(output_spatial_lengths),
-                       std::end(output_spatial_lengths));
-
-    Tensor<InDataType> input_host_result(
-        get_input_host_tensor_descriptor<InLayout>(input_dims, NDimSpatial));
-    Tensor<InDataType> input_device_result(
-        get_input_host_tensor_descriptor<InLayout>(input_dims, NDimSpatial));
-    Tensor<WeiDataType> weights(
-        get_filters_host_tensor_descriptor<WeiLayout>(filter_dims, NDimSpatial));
-    Tensor<OutDataType> output(
-        get_output_host_ensor_descriptor<OutLayout>(output_dims, NDimSpatial));
-
-    std::cout << "input: " << input_host_result.mDesc << std::endl;
-    std::cout << "weights: " << weights.mDesc << std::endl;
-    std::cout << "output: " << output.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        output.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
-        weights.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
-        break;
-    default:
-        output.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
-        weights.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
-    }
-
-    DeviceMem in_device_buf(sizeof(InDataType) * input_device_result.mDesc.GetElementSpace());
-    DeviceMem wei_device_buf(sizeof(WeiDataType) * weights.mDesc.GetElementSpace());
-    DeviceMem out_device_buf(sizeof(OutDataType) * output.mDesc.GetElementSpace());
-
-    out_device_buf.ToDevice(output.mData.data());
-    wei_device_buf.ToDevice(weights.mData.data());
-
-    // reset input to zero
-    in_device_buf.SetZero();
-
-    if(do_verification)
-    {
-        auto RunReference = [&](auto& ref_conv) {
-            auto ref_invoker = ref_conv.MakeInvoker();
-
-            auto ref_argument = ref_conv.MakeArgument(input_host_result,
-                                                      weights,
-                                                      output,
-                                                      conv_filter_strides,
-                                                      conv_filter_dilations,
-                                                      input_left_pads,
-                                                      input_right_pads,
-                                                      InElementOp{},
-                                                      WeiElementOp{},
-                                                      OutElementOp{});
-            ref_invoker.Run(ref_argument);
-        };
-
-        auto ref_conv = ck::tensor_operation::host::ReferenceConvBwdData<InDataType,
-                                                                         WeiDataType,
-                                                                         OutDataType,
-                                                                         AccDataType,
-                                                                         InElementOp,
-                                                                         WeiElementOp,
-                                                                         OutElementOp,
-                                                                         NDimSpatial>();
-        RunReference(ref_conv);
-    }
-
-    // add device Conv instances
-    std::vector<DeviceConvBwdDataNoOpPtr> conv_ptrs;
-    get_device_conv_bwd_data_op_ptr(
-        InDataType{}, WeiDataType{}, OutDataType{}, conv_ptrs, NDimSpatial);
-
-    if(conv_ptrs.size() <= 0)
-    {
-        throw std::runtime_error("wrong! no device Conv instance found");
-    }
-
-    std::string best_conv_name;
-    float best_ave_time   = 0;
-    float best_tflops     = 0;
-    float best_gb_per_sec = 0;
-
-    // profile device Conv instances
-    bool success = true;
-    for(auto& conv_ptr : conv_ptrs)
-    {
-        auto argument_ptr = conv_ptr->MakeArgumentPointer(
-            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
-            static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
-            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-            N,
-            K,
-            C,
-            input_spatial_lengths,
-            filter_spatial_lengths,
-            output_spatial_lengths,
-            conv_filter_strides,
-            conv_filter_dilations,
-            input_left_pads,
-            input_right_pads,
-            in_element_op,
-            wei_element_op,
-            out_element_op);
-
-        auto invoker_ptr = conv_ptr->MakeInvokerPointer();
-
-        if(conv_ptr->IsSupportedArgument(argument_ptr.get()))
-        {
-            std::string conv_name = conv_ptr->GetTypeString();
-
-            float ave_time =
-                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
-
-            std::size_t flop =
-                ck::utils::conv::get_flops(N, C, K, filter_spatial_lengths, output_spatial_lengths);
-            std::size_t num_btype =
-                ck::utils::conv::get_btype<InDataType, WeiDataType, OutDataType>(
-                    N, C, K, input_spatial_lengths, filter_spatial_lengths, output_spatial_lengths);
-
-            float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
-            float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-            std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                      << " GB/s" << std::endl;
-
-            if(tflops > best_tflops)
-            {
-                best_conv_name  = conv_name;
-                best_tflops     = tflops;
-                best_ave_time   = ave_time;
-                best_gb_per_sec = gb_per_sec;
-            }
-
-            if(do_verification)
-            {
-                in_device_buf.FromDevice(input_device_result.mData.data());
-
-                if(!check_out(input_host_result, input_device_result))
-                {
-                    std::cout << "Fail Info: " << conv_ptr->GetTypeString() << std::endl;
-
-                    success = false;
-                }
-                else
-                {
-                    std::cout << "Pass Info: " << conv_ptr->GetTypeString() << std::endl;
-                }
-
-                success = ck::utils::check_err(input_host_result, input_device_result);
-
-                if(do_log)
-                {
-                    std::cout << "in : ";
-                    show_data_nhwc_layout(output);
-                    std::cout << std::endl;
-
-                    std::cout << "wei: ";
-                    show_data_nhwc_layout(weights);
-                    std::cout << std::endl;
-
-                    std::cout << "out_host  : ";
-                    show_data_nhwc_layout(input_host_result);
-                    std::cout << std::endl;
-
-                    std::cout << "out_device: ";
-                    show_data_nhwc_layout(input_device_result);
-                    std::cout << std::endl;
-                }
-            }
-        }
-    }
-
-    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
-              << best_gb_per_sec << " GB/s, " << best_conv_name << std::endl;
-    return success;
-}
-
-} // namespace profiler
-} // namespace ck
diff --git a/profiler/include/profiler/profile_convnd_bwd_weight_impl.hpp b/profiler/include/profiler/profile_convnd_bwd_weight_impl.hpp
deleted file mode 100644
index e37c887a96f9c843abd62ee83f026da06f887daf..0000000000000000000000000000000000000000
--- a/profiler/include/profiler/profile_convnd_bwd_weight_impl.hpp
+++ /dev/null
@@ -1,474 +0,0 @@
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_conv_backward_weight.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/conv_util.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_backward_weight.hpp"
-
-using F16  = ck::half_t;
-using F32  = float;
-using BF16 = ck::bhalf_t;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using DeviceConvndBwdWeightNoOpPtr =
-    DeviceConvBwdWeightPtr<ck::tensor_operation::element_wise::PassThrough,
-                           ck::tensor_operation::element_wise::PassThrough,
-                           ck::tensor_operation::element_wise::PassThrough>;
-
-void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-
-void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-
-void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-namespace ck {
-namespace profiler {
-
-using DeviceConvndBwdWeightNoOpPtr =
-    ck::tensor_operation::device::instance::DeviceConvndBwdWeightNoOpPtr;
-
-template <typename InLayout>
-HostTensorDescriptor get_input_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-
-template <typename WeiLayout>
-HostTensorDescriptor get_filters_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                        int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-
-template <typename OutLayout>
-HostTensorDescriptor get_output_host_ensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-
-template <typename InDataType, typename WeiDataType, typename OutDataType>
-void get_device_conv_bwd_weight_op_ptr(
-    InDataType, WeiDataType, OutDataType, std::vector<DeviceConvndBwdWeightNoOpPtr>&, int)
-{
-    std::cout << "can not find device conv bwd weight" << std::endl;
-    exit(1);
-}
-
-template <>
-void get_device_conv_bwd_weight_op_ptr(
-    F32, F32, F32, std::vector<DeviceConvndBwdWeightNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <>
-void get_device_conv_bwd_weight_op_ptr(
-    F16, F16, F16, std::vector<DeviceConvndBwdWeightNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <>
-void get_device_conv_bwd_weight_op_ptr(
-    BF16, BF16, BF16, std::vector<DeviceConvndBwdWeightNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <typename DataType>
-void show_data_nhwc_layout(Tensor<DataType>& nhwc)
-{
-    std::cout << "[";
-    for(int n = 0; n < ck::type_convert<int>(nhwc.mDesc.GetLengths()[0]); n++)
-    {
-        std::cout << "[";
-        for(int hi = 0; hi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[2]); hi++)
-        {
-            std::cout << "[";
-            for(int wi = 0; wi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[3]); wi++)
-            {
-                std::cout << "[";
-                for(int c = 0; c < ck::type_convert<int>(nhwc.mDesc.GetLengths()[1]); c++)
-                {
-                    std::cout << static_cast<float>(nhwc(n, c, hi, wi)) << "  ";
-                }
-                std::cout << "]";
-            }
-            std::cout << "]";
-        }
-        std::cout << "]";
-    }
-    std::cout << "]";
-}
-
-template <int NDimSpatial,
-          typename InDataType,
-          typename WeiDataType,
-          typename OutDataType,
-          typename InLayout,
-          typename WeiLayout,
-          typename OutLayout>
-bool profile_convnd_bwd_weight_impl(int do_verification,
-                                    int init_method,
-                                    bool do_log,
-                                    bool time_kernel,
-                                    ck::index_t N,
-                                    ck::index_t K,
-                                    ck::index_t C,
-                                    std::vector<ck::index_t> input_spatial_lengths,
-                                    std::vector<ck::index_t> filter_spatial_lengths,
-                                    std::vector<ck::index_t> output_spatial_lengths,
-                                    std::vector<ck::index_t> conv_filter_strides,
-                                    std::vector<ck::index_t> conv_filter_dilations,
-                                    std::vector<ck::index_t> input_left_pads,
-                                    std::vector<ck::index_t> input_right_pads,
-                                    ck::index_t split_k)
-{
-    using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-    using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-    const auto in_element_op  = InElementOp{};
-    const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{};
-
-    std::vector<std::size_t> input_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(C)};
-    input_dims.insert(
-        std::end(input_dims), std::begin(input_spatial_lengths), std::end(input_spatial_lengths));
-
-    std::vector<std::size_t> filter_dims{static_cast<std::size_t>(K), static_cast<std::size_t>(C)};
-    filter_dims.insert(std::end(filter_dims),
-                       std::begin(filter_spatial_lengths),
-                       std::end(filter_spatial_lengths));
-
-    std::vector<std::size_t> output_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(K)};
-    output_dims.insert(std::end(output_dims),
-                       std::begin(output_spatial_lengths),
-                       std::end(output_spatial_lengths));
-
-    Tensor<InDataType> input(get_input_host_tensor_descriptor<InLayout>(input_dims, NDimSpatial));
-    Tensor<WeiDataType> weights_host_result(
-        get_filters_host_tensor_descriptor<WeiLayout>(filter_dims, NDimSpatial));
-    Tensor<WeiDataType> weights_device_result(
-        get_filters_host_tensor_descriptor<WeiLayout>(filter_dims, NDimSpatial));
-    Tensor<OutDataType> output(
-        get_output_host_ensor_descriptor<OutLayout>(output_dims, NDimSpatial));
-
-    std::cout << "input: " << input.mDesc << std::endl;
-    std::cout << "weights: " << weights_host_result.mDesc << std::endl;
-    std::cout << "output: " << output.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        input.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-2, 2});
-        output.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 2});
-        break;
-    default:
-        input.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
-        output.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
-    }
-
-    DeviceMem in_device_buf(sizeof(InDataType) * input.mDesc.GetElementSpace());
-    DeviceMem wei_device_buf(sizeof(WeiDataType) * weights_device_result.mDesc.GetElementSpace());
-    DeviceMem out_device_buf(sizeof(OutDataType) * output.mDesc.GetElementSpace());
-
-    in_device_buf.ToDevice(input.mData.data());
-    out_device_buf.ToDevice(output.mData.data());
-
-    // reset input to zero
-    wei_device_buf.SetZero();
-
-    if(do_verification)
-    {
-        auto RunReference = [&](auto& ref_conv) {
-            auto ref_invoker = ref_conv.MakeInvoker();
-
-            auto ref_argument = ref_conv.MakeArgument(input,
-                                                      weights_host_result,
-                                                      output,
-                                                      conv_filter_strides,
-                                                      conv_filter_dilations,
-                                                      input_left_pads,
-                                                      input_right_pads,
-                                                      InElementOp{},
-                                                      WeiElementOp{},
-                                                      OutElementOp{});
-            ref_invoker.Run(ref_argument);
-        };
-
-        auto ref_conv = ck::tensor_operation::host::ReferenceConvBwdWeight<InDataType,
-                                                                           WeiDataType,
-                                                                           OutDataType,
-                                                                           InElementOp,
-                                                                           WeiElementOp,
-                                                                           OutElementOp,
-                                                                           NDimSpatial>();
-        RunReference(ref_conv);
-    }
-
-    // add device Conv instances
-    std::vector<DeviceConvndBwdWeightNoOpPtr> conv_ptrs;
-    get_device_conv_bwd_weight_op_ptr(
-        InDataType{}, WeiDataType{}, OutDataType{}, conv_ptrs, NDimSpatial);
-
-    if(conv_ptrs.size() <= 0)
-    {
-        throw std::runtime_error("wrong! no device Conv instance found");
-    }
-
-    std::string best_conv_name;
-    float best_ave_time   = 0;
-    float best_tflops     = 0;
-    float best_gb_per_sec = 0;
-
-    // profile device Conv instances
-    bool success = true;
-    for(auto& conv_ptr : conv_ptrs)
-    {
-        // using atomic, so need to reset input, setzero is done in invoker
-        // if(split_k > 1)
-        //{
-        //    wei_device_buf.SetZero();
-        //}
-
-        auto argument_ptr = conv_ptr->MakeArgumentPointer(
-            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
-            static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
-            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-            N,
-            K,
-            C,
-            input_spatial_lengths,
-            filter_spatial_lengths,
-            output_spatial_lengths,
-            conv_filter_strides,
-            conv_filter_dilations,
-            input_left_pads,
-            input_right_pads,
-            in_element_op,
-            wei_element_op,
-            out_element_op,
-            split_k);
-
-        if(!conv_ptr->IsSupportedArgument(argument_ptr.get()))
-        {
-            std::cout << "wrong! device_conv with the specified compilation parameters does "
-                         "not support this Conv problem"
-                      << std::endl;
-            continue;
-        }
-
-        auto invoker_ptr      = conv_ptr->MakeInvokerPointer();
-        std::string conv_name = conv_ptr->GetTypeString();
-        float ave_time        = 0;
-
-        if(std::is_same<InDataType, ck::bhalf_t>::value && split_k > 1)
-        {
-            // alloc work space
-            size_t bwd_weight_workspace_size = conv_ptr->GetWorkSpaceSize(argument_ptr.get());
-            if(bwd_weight_workspace_size <= 0)
-            {
-                printf("wrong work space size\n");
-                exit(1);
-            }
-            DeviceMem wei_work_space_device_buf(bwd_weight_workspace_size);
-            wei_work_space_device_buf.SetZero();
-            conv_ptr->SetWorkSpacePointer(argument_ptr.get(),
-                                          wei_work_space_device_buf.GetDeviceBuffer());
-            ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
-        }
-        else
-        {
-            ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
-        }
-
-        std::size_t flop =
-            ck::utils::conv::get_flops(N, C, K, filter_spatial_lengths, output_spatial_lengths);
-        std::size_t num_btype = ck::utils::conv::get_btype<InDataType, WeiDataType, OutDataType>(
-            N, C, K, input_spatial_lengths, filter_spatial_lengths, output_spatial_lengths);
-
-        float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
-        float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                  << " GB/s" << std::endl;
-
-        if(tflops > best_tflops)
-        {
-            best_conv_name  = conv_name;
-            best_tflops     = tflops;
-            best_ave_time   = ave_time;
-            best_gb_per_sec = gb_per_sec;
-        }
-
-        if(do_verification)
-        {
-            wei_device_buf.FromDevice(weights_device_result.mData.data());
-
-            success = ck::utils::check_err(weights_host_result, weights_device_result);
-
-            if(success == false)
-            {
-                std::cout << "Fail Info: " << conv_ptr->GetTypeString() << std::endl;
-            }
-            else
-            {
-                std::cout << "Pass Info: " << conv_ptr->GetTypeString() << std::endl;
-            }
-
-            if(do_log)
-            {
-                std::cout << "in : ";
-                show_data_nhwc_layout(output);
-                std::cout << std::endl;
-
-                std::cout << "wei: ";
-                show_data_nhwc_layout(weights_host_result);
-                std::cout << std::endl;
-
-                std::cout << "out  : ";
-                show_data_nhwc_layout(input);
-                std::cout << std::endl;
-
-                std::cout << "wei_device: ";
-                show_data_nhwc_layout(weights_device_result);
-                std::cout << std::endl;
-            }
-        }
-    }
-
-    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
-              << best_gb_per_sec << " GB/s, " << best_conv_name << std::endl;
-    return success;
-}
-
-} // namespace profiler
-} // namespace ck
diff --git a/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp b/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
index 7707e16b089ef3058dea988188f80e453fcd4e7d..1fd9c811095a823ce2746ddba63431ccdd77f20b 100644
--- a/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp b/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp
index 3cc2ea3b92624da3db43696c181248e55a5f0256..81b8d8ddbf551f23de68beabd8900b4251c97c2a 100644
--- a/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp b/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp
index d53a6589e0f505ed3beb6a7792c9045c49ee0c68..6f6d881c1e4ef5214fb13d3b7523c2b4c979fd49 100644
--- a/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_add_multiply_impl.hpp b/profiler/include/profiler/profile_gemm_add_multiply_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..25871dfb2ec814fb5d0cda406a31ad8602b9dc16
--- /dev/null
+++ b/profiler/include/profiler/profile_gemm_add_multiply_impl.hpp
@@ -0,0 +1,242 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename D0DataType,
+          typename D1DataType,
+          typename EDataType,
+          typename ALayout,
+          typename BLayout,
+          typename D0Layout,
+          typename D1Layout,
+          typename ELayout>
+bool profile_gemm_add_multiply_impl(int do_verification,
+                                    int init_method,
+                                    bool /*do_log*/,
+                                    bool time_kernel,
+                                    int M,
+                                    int N,
+                                    int K,
+                                    int StrideA,
+                                    int StrideB,
+                                    int StrideD0,
+                                    int StrideD1,
+                                    int StrideE)
+{
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(is_same<decltype(layout), tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<D0DataType> d0_m_n(f_host_tensor_descriptor(M, N, StrideD0, D0Layout{}));
+    Tensor<D1DataType> d1_m_n(f_host_tensor_descriptor(M, N, StrideD1, D1Layout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d0_m_n: " << d0_m_n.mDesc << std::endl;
+    std::cout << "d1_m_n: " << d1_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_device_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d0_m_n.GenerateTensorValue(GeneratorTensor_2<D0DataType>{-5, 5});
+        d1_m_n.GenerateTensorValue(GeneratorTensor_2<D1DataType>{-1, 1});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d0_m_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
+        d1_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
+    }
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+    using AddMultiply = ck::tensor_operation::element_wise::AddMultiply;
+
+    using AElementOp   = PassThrough;
+    using BElementOp   = PassThrough;
+    using CDEElementOp = AddMultiply;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceGemmMultipleD<ALayout,
+                                                          BLayout,
+                                                          ck::Tuple<D0Layout, D1Layout>,
+                                                          ELayout,
+                                                          ADataType,
+                                                          BDataType,
+                                                          ck::Tuple<D0DataType, D1DataType>,
+                                                          EDataType,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    // run reference
+    if(do_verification)
+    {
+        Tensor<AccDataType> c_m_n({M, N});
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                AccDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d0_m_n(m, n), d1_m_n(m, n));
+            }
+        }
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d0_m_n_device_buf(sizeof(D0DataType) * d0_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem d1_m_n_device_buf(sizeof(D1DataType) * d1_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d0_m_n_device_buf.ToDevice(d0_m_n.mData.data());
+    d1_m_n_device_buf.ToDevice(d1_m_n.mData.data());
+
+    std::string best_op_name;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    bool pass = true;
+
+    // profile device operation instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b_device_buf.GetDeviceBuffer(),
+            std::array<const void*, 2>{d0_m_n_device_buf.GetDeviceBuffer(),
+                                       d1_m_n_device_buf.GetDeviceBuffer()},
+            e_device_buf.GetDeviceBuffer(),
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            std::array<ck::index_t, 2>{StrideD0, StrideD1},
+            StrideE,
+            a_element_op,
+            b_element_op,
+            cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init E to zero before profiling a kernel
+            e_device_buf.SetZero();
+
+            float ave_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop = std::size_t(2) * M * N * K;
+
+            std::size_t num_btype =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+                pass = pass && ck::utils::check_err(e_m_n_device_result, e_m_n_host_result);
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp b/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..4c3d0a045054841b3f88a9c85ba9eaa8fd2538b7
--- /dev/null
+++ b/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
@@ -0,0 +1,346 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename D0DataType,
+          typename D1DataType,
+          typename EMeanVarDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename HDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp,
+          typename HElementOp>
+void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
+                         const Tensor<ADataType>& a_m_k,
+                         const Tensor<BDataType>& b_k_n,
+                         const Tensor<D0DataType>& d0_m_n,
+                         const Tensor<D1DataType>& d1_m_n,
+                         const Tensor<GammaDataType>& gamma_n,
+                         const Tensor<BetaDataType>& beta_n,
+                         AElementOp a_element_op,
+                         BElementOp b_element_op,
+                         CDEElementOp cde_element_op,
+                         HElementOp h_element_op,
+                         int M,
+                         int N,
+                         AccDataType epsilon = 1e-5)
+{
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    using ReferenceGemm = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                    BDataType,
+                                                                    AccDataType,
+                                                                    AccDataType,
+                                                                    AElementOp,
+                                                                    BElementOp,
+                                                                    PassThrough>;
+
+    using ReferenceLayernorm = ck::tensor_operation::host::ReferenceLayernorm<EMeanVarDataType,
+                                                                              GammaDataType,
+                                                                              BetaDataType,
+                                                                              HDataType,
+                                                                              AccDataType,
+                                                                              HElementOp,
+                                                                              2,
+                                                                              1>;
+
+    Tensor<EMeanVarDataType> e_m_n(HostTensorDescriptor{M, N});
+    Tensor<AccDataType> c_m_n(HostTensorDescriptor{M, N});
+
+    auto ref_gemm         = ReferenceGemm{};
+    auto ref_gemm_invoker = ref_gemm.MakeInvoker();
+
+    auto ref_gemm_argument =
+        ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
+
+    ref_gemm_invoker.Run(ref_gemm_argument);
+
+    for(int n = 0; n < N; ++n)
+    {
+        for(int m = 0; m < M; ++m)
+        {
+            AccDataType e  = static_cast<AccDataType>(e_m_n(m, n));
+            AccDataType d0 = static_cast<AccDataType>(d0_m_n(m, n));
+            AccDataType d1 = static_cast<AccDataType>(d1_m_n(m, n));
+            cde_element_op(e, c_m_n(m, n), d0, d1);
+            e_m_n(m, n) = static_cast<EMeanVarDataType>(e);
+        }
+    }
+
+    ReferenceLayernorm ref_layernorm;
+    auto ref_layernorm_invoker = ref_layernorm.MakeInvoker();
+
+    auto ref_layernorm_argument = ref_layernorm.MakeArgument(
+        e_m_n, gamma_n, beta_n, h_m_n, h_element_op, {M, N}, {1}, epsilon);
+    ref_layernorm_invoker.Run(ref_layernorm_argument);
+}
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename D0DataType,
+          typename D1DataType,
+          typename EMeanVarDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename HDataType,
+          typename ALayout,
+          typename BLayout,
+          typename D0Layout,
+          typename D1Layout,
+          typename HLayout>
+bool profile_gemm_add_relu_add_layernorm_impl(int do_verification,
+                                              int init_method,
+                                              bool /*do_log*/,
+                                              bool time_kernel,
+                                              int M,
+                                              int N,
+                                              int K,
+                                              int StrideA,
+                                              int StrideB,
+                                              int StrideD0,
+                                              int StrideD1,
+                                              int StrideH,
+                                              AccDataType epsilon = 1e-5)
+{
+    auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
+        return HostTensorDescriptor({len}, {stride});
+    };
+
+    auto f_host_tensor_descriptor2d =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if constexpr(std::is_same<decltype(layout), tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor2d(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor2d(K, N, StrideB, BLayout{}));
+    Tensor<D1DataType> d0_m_n(f_host_tensor_descriptor2d(M, N, StrideD0, D0Layout{}));
+    Tensor<D1DataType> d1_m_n(f_host_tensor_descriptor2d(M, N, StrideD1, D1Layout{}));
+    Tensor<GammaDataType> gamma_n(f_host_tensor_descriptor1d(N, 1));
+    Tensor<BetaDataType> beta_n(f_host_tensor_descriptor1d(N, 1));
+    Tensor<HDataType> h_m_n(f_host_tensor_descriptor2d(M, N, StrideH, HLayout{}));
+    Tensor<HDataType> h_m_n_host(f_host_tensor_descriptor2d(M, N, StrideH, HLayout{}));
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{-1, 1});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-1, 1});
+        d0_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{-1, 1});
+        d1_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{-1, 1});
+        gamma_n.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{-1, 1});
+        beta_n.GenerateTensorValue(GeneratorTensor_3<BetaDataType>{-1, 1});
+        break;
+    }
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+    using AddReluAdd  = ck::tensor_operation::element_wise::AddReluAdd;
+
+    using AElementOp   = PassThrough;
+    using BElementOp   = PassThrough;
+    using CDEElementOp = AddReluAdd;
+    using HElementOp   = PassThrough;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+    const auto h_element_op   = HElementOp{};
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleDLayernorm<
+        ALayout,
+        BLayout,
+        ck::Tuple<D0Layout, D1Layout>,
+        HLayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<D0DataType, D1DataType>,
+        GammaDataType,
+        BetaDataType,
+        HDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::AddReluAdd,
+        ck::tensor_operation::element_wise::PassThrough>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    // run reference
+    if(do_verification)
+    {
+        host_gemm_layernorm<ADataType,
+                            BDataType,
+                            AccDataType,
+                            D0DataType,
+                            D1DataType,
+                            EMeanVarDataType,
+                            GammaDataType,
+                            BetaDataType,
+                            HDataType>(h_m_n_host,
+                                       a_m_k,
+                                       b_k_n,
+                                       d0_m_n,
+                                       d1_m_n,
+                                       gamma_n,
+                                       beta_n,
+                                       a_element_op,
+                                       b_element_op,
+                                       cde_element_op,
+                                       h_element_op,
+                                       M,
+                                       N,
+                                       epsilon);
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d0_m_n_device_buf(sizeof(D0DataType) * d0_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem d1_m_n_device_buf(sizeof(D1DataType) * d1_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem gamma_device_buf(sizeof(GammaDataType) * gamma_n.mDesc.GetElementSpaceSize());
+    DeviceMem beta_device_buf(sizeof(BetaDataType) * beta_n.mDesc.GetElementSpaceSize());
+    DeviceMem h_device_buf(sizeof(HDataType) * h_m_n.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d0_m_n_device_buf.ToDevice(d0_m_n.mData.data());
+    d1_m_n_device_buf.ToDevice(d1_m_n.mData.data());
+    gamma_device_buf.ToDevice(gamma_n.mData.data());
+    beta_device_buf.ToDevice(beta_n.mData.data());
+
+    std::string best_op_name;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    bool pass      = true;
+    int num_kernel = 0;
+
+    // profile device operation instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b_device_buf.GetDeviceBuffer(),
+            {d0_m_n_device_buf.GetDeviceBuffer(), d1_m_n_device_buf.GetDeviceBuffer()},
+            gamma_device_buf.GetDeviceBuffer(),
+            beta_device_buf.GetDeviceBuffer(),
+            h_device_buf.GetDeviceBuffer(),
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            {StrideD0, StrideD1},
+            StrideH,
+            epsilon,
+            a_element_op,
+            b_element_op,
+            cde_element_op,
+            h_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            ++num_kernel;
+
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+            DeviceMem workspace_dev(workspace_sz);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
+            // re-init E to zero before profiling a kernel
+            h_device_buf.SetZero();
+
+            float ave_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t num_byte =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                (sizeof(D0DataType) + sizeof(D1DataType) + sizeof(HDataType)) * M * N +
+                (sizeof(GammaDataType) + sizeof(BetaDataType)) * N;
+
+            float gb_per_sec = num_byte / 1.E6 / ave_time;
+
+            if(time_kernel)
+                std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec
+                          << " GB/s, " << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                h_device_buf.FromDevice(h_m_n.mData.data());
+
+                pass = pass && ck::utils::check_err(
+                                   h_m_n, h_m_n_host, "Error: Incorrect results h_m_n", 1e-2, 1e-2);
+            }
+        }
+        else
+        {
+            if(time_kernel)
+                std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        pass = false;
+    }
+    else
+    {
+        if(time_kernel)
+            std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                      << best_op_name << std::endl;
+    }
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp b/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp
index b4ec78cdf37659b6e838b10a8c235fdd6a0c9c64..c0ffea8a326c8bbc95a9b2d2e46da651089d807a 100644
--- a/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_bilinear_impl.hpp b/profiler/include/profiler/profile_gemm_bilinear_impl.hpp
index 31bae281c45b2e4e9bbdd8068fd3b084270c45ad..b540e938b5b5ab5798a6a18f86e452caf819c75e 100644
--- a/profiler/include/profiler/profile_gemm_bilinear_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_bilinear_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp b/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp
index f9a544c044f4165fbee8cab9ba10a10616f27e22..3893f8cdc7cfbf957fe28cd2b019092b59c88029 100644
--- a/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_impl.hpp b/profiler/include/profiler/profile_gemm_impl.hpp
index 9b164104b505d3d3912206b5fa1a6ee6aeb2ef48..eaab5dbcc2c8d3a6d5e1ea5d59c1c292f696d16d 100644
--- a/profiler/include/profiler/profile_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_reduce_impl.hpp b/profiler/include/profiler/profile_gemm_reduce_impl.hpp
index 370121a3ccff382539ee34994a05b4b035db5512..ff801e8afd3731e62bd6edb545cce87bbddeae74 100644
--- a/profiler/include/profiler/profile_gemm_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_reduce_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index e5d5f8765e1b25e67490da7444f31ac38c7aca27..6ffa31678f837a670f750dff8c6288d39cafb20a 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -72,8 +72,8 @@ bool profile_gemm_splitk_impl(int do_verification,
     {
     case 0: break;
     case 1:
-        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-1, 2});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-1, 2});
         break;
     default:
         a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
@@ -94,7 +94,7 @@ bool profile_gemm_splitk_impl(int do_verification,
 
     a_device_buf.ToDevice(a_m_k.mData.data());
     b_device_buf.ToDevice(b_k_n.mData.data());
-    c_device_buf.ToDevice(c_m_n_device_result.mData.data());
+    c_device_buf.SetZero();
 
     using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK<ALayout,
                                                                     BLayout,
@@ -246,9 +246,9 @@ bool profile_gemm_splitk_impl(int do_verification,
     }
 
     std::cout << " M = " << M << " N = " << N << " K = " << K << " StrideA = " << StrideA
-              << " StrideB = " << StrideB << " StrideC = " << StrideC << " : " << best_ave_time
-              << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec << " GB/s, "
-              << best_op_name << std::endl;
+              << " StrideB = " << StrideB << " StrideC = " << StrideC << " KBatch = " << KBatch
+              << " : " << best_ave_time << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec
+              << " GB/s, " << best_op_name << std::endl;
 
     return pass;
 }
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..93d3430bba30448b4cd1e62d402bc251e1116191
--- /dev/null
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp
@@ -0,0 +1,257 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <ck::index_t NDimSpatial,
+          typename OutLayout,
+          typename WeiLayout,
+          typename InLayout,
+          typename OutDataType,
+          typename WeiDataType,
+          typename InDataType>
+bool profile_grouped_conv_bwd_data_impl(int do_verification,
+                                        int init_method,
+                                        bool do_log,
+                                        bool time_kernel,
+                                        const ck::utils::conv::ConvParam& conv_param)
+{
+    using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
+
+    const auto out_element_op = OutElementOp{};
+    const auto wei_element_op = WeiElementOp{};
+    const auto in_element_op  = InElementOp{};
+
+    const auto out_g_n_k_wos_desc =
+        ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
+
+    const auto wei_g_k_c_xs_desc =
+        ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(conv_param);
+
+    const auto in_g_n_c_wis_desc =
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
+
+    Tensor<OutDataType> out(out_g_n_k_wos_desc);
+    Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<InDataType> in_host(in_g_n_c_wis_desc);
+    Tensor<InDataType> in_device(in_g_n_c_wis_desc);
+
+    std::cout << "out: " << out.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "in: " << in_host.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
+        wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
+        break;
+    case 2:
+        out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{0.0, 1.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
+        break;
+    default:
+        out.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
+        wei.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
+    }
+
+    DeviceMem out_device_buf(sizeof(OutDataType) * out.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem in_device_buf(sizeof(InDataType) * in_device.mDesc.GetElementSpaceSize());
+
+    out_device_buf.ToDevice(out.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+
+    // reset input to zero
+    in_device_buf.SetZero();
+
+    if(do_verification)
+    {
+        auto ref_conv = ck::tensor_operation::host::ReferenceConvBwdData<NDimSpatial,
+                                                                         InDataType,
+                                                                         WeiDataType,
+                                                                         OutDataType,
+                                                                         InElementOp,
+                                                                         WeiElementOp,
+                                                                         OutElementOp>();
+
+        auto ref_invoker = ref_conv.MakeInvoker();
+
+        in_host.SetZero();
+
+        auto ref_argument = ref_conv.MakeArgument(in_host,
+                                                  wei,
+                                                  out,
+                                                  conv_param.conv_filter_strides_,
+                                                  conv_param.conv_filter_dilations_,
+                                                  conv_param.input_left_pads_,
+                                                  conv_param.input_right_pads_,
+                                                  out_element_op,
+                                                  wei_element_op,
+                                                  in_element_op);
+
+        ref_invoker.Run(ref_argument);
+    }
+
+    std::string best_op_name;
+    float best_avg_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    bool pass = true;
+
+    auto run_impl = [&](auto& op_ptr, auto& argument_ptr) {
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init output to zero before profiling next kernel
+            in_device_buf.SetZero();
+
+            std::string op_name = op_ptr->GetTypeString();
+
+            auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+            float avg_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop      = conv_param.GetFlops();
+            std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+            float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
+
+            float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                in_device_buf.FromDevice(in_device.mData.data());
+
+                pass = pass & ck::utils::check_err(in_device, in_host);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "output : ", out.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "weight: ", wei.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "in_host  : ", in_host.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "in_device: ", in_device.mData, ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    };
+
+    // do GEMM
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NDimSpatial,
+                                                                                     OutLayout,
+                                                                                     WeiLayout,
+                                                                                     ck::Tuple<>,
+                                                                                     InLayout,
+                                                                                     OutDataType,
+                                                                                     WeiDataType,
+                                                                                     ck::Tuple<>,
+                                                                                     InDataType,
+                                                                                     OutElementOp,
+                                                                                     WeiElementOp,
+                                                                                     InElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> out_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> in_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
+    copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+                                        static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
+                                        {},
+                                        static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+                                        out_lengths,
+                                        out_strides,
+                                        wei_lengths,
+                                        wei_strides,
+                                        {},
+                                        {},
+                                        in_lengths,
+                                        in_strides,
+                                        conv_filter_strides,
+                                        conv_filter_dilations,
+                                        input_left_pads,
+                                        input_right_pads,
+                                        out_element_op,
+                                        wei_element_op,
+                                        in_element_op);
+
+        run_impl(op_ptr, argument_ptr);
+    }
+
+    std::cout << "Best configuration parameters:"
+              << "\nname: " << best_op_name << "\navg_time: " << best_avg_time
+              << "\ntflops: " << best_tflops << "\nGB/s: " << best_gb_per_sec << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
index 4f9aa98376c9e56058f04938f8171ac93c230e3e..3a4295eeb32d43bcf1c1e99130679bde84f191bc 100644
--- a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -139,6 +139,8 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> output_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -149,6 +151,8 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
     range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
     range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
     range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+    range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+    range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
     range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
     range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
     range_copy(conv_param.input_left_pads_, begin(input_left_pads));
@@ -167,6 +171,8 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
                                         input_spatial_lengths,
                                         filter_spatial_lengths,
                                         output_spatial_lengths,
+                                        input_strides,
+                                        output_strides,
                                         conv_filter_strides,
                                         conv_filter_dilations,
                                         input_left_pads,
diff --git a/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp b/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
index b201a2ed331cbf4a0e3951d8d69027f018af09f6..9fadfe96992a8d0374e784d1a95c4ac7989562bd 100644
--- a/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_grouped_gemm_fastgelu_impl.hpp b/profiler/include/profiler/profile_grouped_gemm_fastgelu_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..f05b13b7495dfefee263f51d5404dda3dbc598dc
--- /dev/null
+++ b/profiler/include/profiler/profile_grouped_gemm_fastgelu_impl.hpp
@@ -0,0 +1,280 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_fastgelu.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/fill.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout>
+bool profile_grouped_gemm_fastgelu_impl(int do_verification,
+                                        int init_method,
+                                        bool do_log,
+                                        bool time_kernel,
+                                        const std::vector<int>& Ms,
+                                        const std::vector<int>& Ns,
+                                        const std::vector<int>& Ks,
+                                        const std::vector<int>& StrideAs,
+                                        const std::vector<int>& StrideBs,
+                                        const std::vector<int>& StrideCs)
+{
+
+    bool pass = true;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(is_same<decltype(layout), tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    std::size_t group_count = Ms.size();
+
+    if(!(group_count == Ns.size() && group_count == Ks.size() && group_count == StrideAs.size() &&
+         group_count == StrideBs.size() && group_count == StrideCs.size()))
+    {
+        throw std::runtime_error("wrong! inconsistent M/N/Ks, StrideA/B/Cs size\n");
+    }
+
+    std::vector<Tensor<ADataType>> a_m_k;
+    std::vector<Tensor<BDataType>> b_k_n;
+    std::vector<Tensor<CDataType>> c_m_n_device_results;
+
+    for(std::size_t i = 0; i < group_count; i++)
+    {
+        a_m_k.push_back(
+            Tensor<ADataType>(f_host_tensor_descriptor(Ms[i], Ks[i], StrideAs[i], ALayout{})));
+        b_k_n.push_back(
+            Tensor<BDataType>(f_host_tensor_descriptor(Ks[i], Ns[i], StrideBs[i], BLayout{})));
+
+        c_m_n_device_results.push_back(
+            Tensor<CDataType>(f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{})));
+
+        std::cout << "group: " << i << " a_m_k[" << i << "]:" << a_m_k[i].mDesc << ", b_k_n[" << i
+                  << "]:" << b_k_n[i].mDesc << ", c_m_n_device_results[" << i
+                  << "]:" << c_m_n_device_results[i].mDesc << std::endl;
+
+        switch(init_method)
+        {
+        case 0: break;
+        case 1:
+            ck::utils::FillUniformDistributionIntegerValue<ADataType>{}(a_m_k[i]);
+            ck::utils::FillUniformDistributionIntegerValue<BDataType>{}(b_k_n[i]);
+            break;
+        default:
+            ck::utils::FillUniformDistribution<ADataType>{0.0, 1.0}(a_m_k[i]);
+            ck::utils::FillUniformDistribution<BDataType>{-0.5, 0.5}(b_k_n[i]);
+        }
+
+        ck::utils::FillConstant<CDataType>{}(c_m_n_device_results[i]);
+    }
+
+    using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using CElementOp = ck::tensor_operation::element_wise::FastGelu;
+
+    const auto a_element_op = AElementOp{};
+    const auto b_element_op = BElementOp{};
+    const auto c_element_op = CElementOp{};
+
+    using DeviceMemPtr = std::unique_ptr<DeviceMem>;
+    std::vector<DeviceMemPtr> a_device_buf, b_device_buf, c_device_buf;
+
+    a_device_buf.reserve(group_count);
+    b_device_buf.reserve(group_count);
+    c_device_buf.reserve(group_count);
+
+    std::vector<const void*> p_a, p_b;
+    std::vector<void*> p_c;
+
+    p_a.reserve(group_count);
+    p_b.reserve(group_count);
+    p_c.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+
+    gemm_descs.reserve(group_count);
+
+    for(std::size_t i = 0; i < group_count; i++)
+    {
+        a_device_buf.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(ADataType) * a_m_k[i].mDesc.GetElementSpaceSize()));
+        b_device_buf.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(BDataType) * b_k_n[i].mDesc.GetElementSpaceSize()));
+        c_device_buf.emplace_back(std::make_unique<DeviceMem>(
+            sizeof(CDataType) * c_m_n_device_results[i].mDesc.GetElementSpaceSize()));
+
+        a_device_buf[i]->ToDevice(a_m_k[i].mData.data());
+        b_device_buf[i]->ToDevice(b_k_n[i].mData.data());
+        c_device_buf[i]->SetZero();
+
+        gemm_descs.push_back({Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
+
+        p_a.push_back(a_device_buf[i]->GetDeviceBuffer());
+        p_b.push_back(b_device_buf[i]->GetDeviceBuffer());
+        p_c.push_back(c_device_buf[i]->GetDeviceBuffer());
+    }
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedGemm<ALayout,
+                                                                     BLayout,
+                                                                     ck::Tuple<>,
+                                                                     CLayout,
+                                                                     ADataType,
+                                                                     BDataType,
+                                                                     ck::Tuple<>,
+                                                                     CDataType,
+                                                                     AElementOp,
+                                                                     BElementOp,
+                                                                     CElementOp>;
+
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    if(op_ptrs.size() <= 0)
+    {
+        throw std::runtime_error("wrong! no device GEMM instance found");
+    }
+
+    std::string best_gemm_name;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    auto p_ds = std::vector<std::array<const void*, 0>>{};
+
+    // profile device GEMM instances
+    for(auto& gemm_ptr : op_ptrs)
+    {
+        auto argument_ptr = gemm_ptr->MakeArgumentPointer(
+            p_a, p_b, p_ds, p_c, gemm_descs, a_element_op, b_element_op, c_element_op);
+
+        auto invoker_ptr = gemm_ptr->MakeInvokerPointer();
+        DeviceMem gemm_desc_workspace(gemm_ptr->GetWorkSpaceSize(argument_ptr.get()));
+        gemm_ptr->SetWorkSpacePointer(argument_ptr.get(), gemm_desc_workspace.GetDeviceBuffer());
+
+        if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            std::string gemm_name = gemm_ptr->GetTypeString();
+
+            float ave_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop = 0, num_btype = 0;
+            for(std::size_t i = 0; i < gemm_descs.size(); i++)
+            {
+                flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
+                num_btype += sizeof(ADataType) * Ms[i] * Ks[i] + sizeof(BDataType) * Ks[i] * Ns[i] +
+                             sizeof(CDataType) * Ms[i] * Ns[i];
+            }
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << gemm_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_gemm_name  = gemm_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                for(std::size_t i = 0; i < gemm_descs.size(); i++)
+                {
+
+                    c_device_buf[i]->FromDevice(c_m_n_device_results[i].mData.data());
+                    Tensor<CDataType> c_m_n_host_result(
+                        f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{}));
+
+                    using ReferenceGemmInstance =
+                        ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                  BDataType,
+                                                                  CDataType,
+                                                                  AccDataType,
+                                                                  AElementOp,
+                                                                  BElementOp,
+                                                                  CElementOp>;
+
+                    auto ref_gemm     = ReferenceGemmInstance{};
+                    auto ref_invoker  = ref_gemm.MakeInvoker();
+                    auto ref_argument = ref_gemm.MakeArgument(a_m_k[i],
+                                                              b_k_n[i],
+                                                              c_m_n_host_result,
+                                                              a_element_op,
+                                                              b_element_op,
+                                                              c_element_op);
+
+                    ref_invoker.Run(ref_argument);
+
+                    bool group_pass =
+                        ck::utils::check_err(c_m_n_device_results[i], c_m_n_host_result);
+                    pass = pass && group_pass;
+
+                    std::cout << "group: " << i << " verification result: " << std::boolalpha
+                              << group_pass << std::endl;
+
+                    if(do_log)
+                    {
+                        LogRangeAsType<float>(std::cout << "a : ", a_m_k[i].mData, ",")
+                            << std::endl;
+                        LogRangeAsType<float>(std::cout << "b: ", b_k_n[i].mData, ",") << std::endl;
+                        LogRangeAsType<float>(
+                            std::cout << "c_device: ", c_m_n_device_results[i].mData, ",")
+                            << std::endl;
+                        LogRangeAsType<float>(
+                            std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
+                            << std::endl;
+                    }
+                }
+            }
+        }
+        else
+        {
+            std::cout << "does not support this GEMM problem" << std::endl;
+        }
+    }
+
+    if(do_verification)
+    {
+        std::cout << "Verification: " << (pass ? "SUCCESS" : "FAILURE") << std::endl;
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_gemm_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_grouped_gemm_impl.hpp b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
index 04f94a0f244bf79f39f10470b09230a9f8a5fc3a..09a651d77c994e69635cb4891fa0cd55ae52791b 100644
--- a/profiler/include/profiler/profile_grouped_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -8,6 +8,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp"
@@ -18,6 +19,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/literals.hpp"
+#include "ck/library/utility/fill.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 namespace ck {
@@ -39,9 +41,9 @@ bool profile_grouped_gemm_impl(int do_verification,
                                const std::vector<int>& Ks,
                                const std::vector<int>& StrideAs,
                                const std::vector<int>& StrideBs,
-                               const std::vector<int>& StrideCs)
+                               const std::vector<int>& StrideCs,
+                               int kbatch = 1)
 {
-
     bool pass = true;
 
     auto f_host_tensor_descriptor =
@@ -79,11 +81,11 @@ bool profile_grouped_gemm_impl(int do_verification,
 
         c_m_n_device_results.push_back(
             Tensor<CDataType>(f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{})));
-
+#if DEBUG_LOG
         std::cout << "group: " << i << " a_m_k[" << i << "]:" << a_m_k[i].mDesc << ", b_k_n[" << i
                   << "]:" << b_k_n[i].mDesc << ", c_m_n_device_results[" << i
                   << "]:" << c_m_n_device_results[i].mDesc << std::endl;
-
+#endif // DEBUG_LOG
         std::size_t num_thread = 1;
         switch(init_method)
         {
@@ -96,8 +98,6 @@ bool profile_grouped_gemm_impl(int do_verification,
             a_m_k[i].GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0}, num_thread);
             b_k_n[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}, num_thread);
         }
-
-        c_m_n_device_results[i].GenerateTensorValue(GeneratorTensor_0<CDataType>{}, num_thread);
     }
 
     using AElementOp = ck::tensor_operation::element_wise::PassThrough;
@@ -108,11 +108,6 @@ bool profile_grouped_gemm_impl(int do_verification,
     const auto b_element_op = BElementOp{};
     const auto c_element_op = CElementOp{};
 
-    // if(do_verification)
-    // {
-
-    // }
-
     using DeviceMemPtr = std::unique_ptr<DeviceMem>;
     std::vector<DeviceMemPtr> a_device_buf, b_device_buf, c_device_buf;
 
@@ -137,13 +132,12 @@ bool profile_grouped_gemm_impl(int do_verification,
             std::make_unique<DeviceMem>(sizeof(ADataType) * a_m_k[i].mDesc.GetElementSpaceSize()));
         b_device_buf.emplace_back(
             std::make_unique<DeviceMem>(sizeof(BDataType) * b_k_n[i].mDesc.GetElementSpaceSize()));
-
         c_device_buf.emplace_back(std::make_unique<DeviceMem>(
             sizeof(CDataType) * c_m_n_device_results[i].mDesc.GetElementSpaceSize()));
 
         a_device_buf[i]->ToDevice(a_m_k[i].mData.data());
         b_device_buf[i]->ToDevice(b_k_n[i].mData.data());
-        c_device_buf[i]->ToDevice(c_m_n_device_results[i].mData.data());
+        c_device_buf[i]->SetZero();
 
         gemm_descs.push_back({Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
 
@@ -197,43 +191,71 @@ bool profile_grouped_gemm_impl(int do_verification,
         DeviceMem gemm_desc_workspace(gemm_ptr->GetWorkSpaceSize(argument_ptr.get()));
 
         gemm_ptr->SetWorkSpacePointer(argument_ptr.get(), gemm_desc_workspace.GetDeviceBuffer());
+        std::string gemm_name = gemm_ptr->GetTypeString();
+
+        if(kbatch > 1)
+        {
+            using DeviceOpSplitK =
+                ck::tensor_operation::device::DeviceGroupedGemmSplitK<ALayout,
+                                                                      BLayout,
+                                                                      ck::Tuple<>,
+                                                                      CLayout,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      ck::Tuple<>,
+                                                                      CDataType,
+                                                                      AElementOp,
+                                                                      BElementOp,
+                                                                      CElementOp>;
+
+            if(dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get()) != nullptr)
+            {
+                dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get())
+                    ->SetKBatchSize(argument_ptr.get(), kbatch);
+            }
+        }
 
         if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
         {
-            std::string gemm_name = gemm_ptr->GetTypeString();
 
             float ave_time =
                 invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
 
-            std::size_t flop = 0, num_btype = 0;
-            for(std::size_t i = 0; i < gemm_descs.size(); i++)
+            if(time_kernel)
             {
-                flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
+                std::size_t flop = 0, num_btype = 0;
+                for(std::size_t i = 0; i < gemm_descs.size(); i++)
+                {
+                    flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
 
-                num_btype += sizeof(ADataType) * Ms[i] * Ks[i] + sizeof(BDataType) * Ks[i] * Ns[i] +
-                             sizeof(CDataType) * Ms[i] * Ns[i];
-            }
+                    num_btype += sizeof(ADataType) * Ms[i] * Ks[i] +
+                                 sizeof(BDataType) * Ks[i] * Ns[i] +
+                                 sizeof(CDataType) * Ms[i] * Ns[i];
+                }
 
-            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+                float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
 
-            float gb_per_sec = num_btype / 1.E6 / ave_time;
-            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
-                      << gb_per_sec << " GB/s, " << gemm_name << std::endl;
+                float gb_per_sec = num_btype / 1.E6 / ave_time;
+                std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
+                          << " TFlops, " << gb_per_sec << " GB/s, " << gemm_name << std::endl;
 
-            if(tflops > best_tflops)
-            {
-                best_gemm_name  = gemm_name;
-                best_tflops     = tflops;
-                best_ave_time   = ave_time;
-                best_gb_per_sec = gb_per_sec;
+                if(tflops > best_tflops)
+                {
+                    best_gemm_name  = gemm_name;
+                    best_tflops     = tflops;
+                    best_ave_time   = ave_time;
+                    best_gb_per_sec = gb_per_sec;
+                }
             }
 
             if(do_verification)
             {
+                bool instance_pass = true;
                 for(std::size_t i = 0; i < gemm_descs.size(); i++)
                 {
 
                     c_device_buf[i]->FromDevice(c_m_n_device_results[i].mData.data());
+                    c_device_buf[i]->SetZero();
 
                     Tensor<CDataType> c_m_n_host_result(
                         f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{}));
@@ -258,7 +280,20 @@ bool profile_grouped_gemm_impl(int do_verification,
                                                               c_element_op);
 
                     ref_invoker.Run(ref_argument);
-                    pass = pass && ck::utils::check_err(c_m_n_device_results[i], c_m_n_host_result);
+                    if(std::is_same_v<CDataType, ck::half_t> && kbatch > 1)
+                    {
+                        instance_pass =
+                            instance_pass && ck::utils::check_err(c_m_n_device_results[i],
+                                                                  c_m_n_host_result,
+                                                                  "Error: Incorrect results!",
+                                                                  0.06);
+                    }
+                    else
+                    {
+                        instance_pass =
+                            instance_pass &&
+                            ck::utils::check_err(c_m_n_device_results[i], c_m_n_host_result);
+                    }
 
                     if(do_log)
                     {
@@ -273,19 +308,28 @@ bool profile_grouped_gemm_impl(int do_verification,
                             << std::endl;
                     }
                 }
+
+                std::cout << "Instance: " << gemm_name << " verification "
+                          << (instance_pass ? "SUCCEED" : "FAILED") << std::endl;
+
+                pass = pass && instance_pass;
             }
         }
         else
         {
-            std::cout << "does not support this GEMM problem" << std::endl;
+            std::cout << "Instance: " << gemm_name << ", does not support this GEMM problem"
+                      << std::endl;
         }
     }
 
-    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
-              << best_gb_per_sec << " GB/s, " << best_gemm_name << std::endl;
+    if(time_kernel)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+                  << best_gb_per_sec << " GB/s, " << best_gemm_name << std::endl;
+    }
 
     return pass;
-} // namespace profiler
+}
 
 } // namespace profiler
 } // namespace ck
diff --git a/profiler/include/profiler/profile_groupnorm_impl.hpp b/profiler/include/profiler/profile_groupnorm_impl.hpp
index 81fec5590a8463ed68e1cf9797800f99e3fbb3c3..ebefe3dad42a91549b70ea26e171c40b78561eeb 100644
--- a/profiler/include/profiler/profile_groupnorm_impl.hpp
+++ b/profiler/include/profiler/profile_groupnorm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -190,9 +190,9 @@ bool profile_groupnorm_impl(int do_verification,
 
     if(time_kernel)
     {
-        LogRange(std::cout << "length = ", length, ",") << ", ";
-        std::cout << "num_kernel = " << num_kernel << ", best perf = " << best_avg_time << " ms, "
-                  << best_gb_per_sec << " GB/s, " << best_instance_name << std::endl;
+        LogRange(std::cout << "length = ", length, ",") << std::endl;
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
     }
 
     if(num_kernel == 0)
diff --git a/profiler/include/profiler/profile_layernorm_impl.hpp b/profiler/include/profiler/profile_layernorm_impl.hpp
index eb21d4a586090937bf35491eda492348331503a4..2d87c8c8fe98bb32c84d3ffb851e5bd9a4f51525 100644
--- a/profiler/include/profiler/profile_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_layernorm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -19,7 +19,7 @@ namespace profiler {
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
-          typename AccDataType,
+          typename ComputeDataType,
           typename YDataType,
           index_t Rank>
 bool profile_layernorm_impl(int do_verification,
@@ -86,7 +86,7 @@ bool profile_layernorm_impl(int do_verification,
     using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
                                                                        GammaDataType,
                                                                        BetaDataType,
-                                                                       AccDataType,
+                                                                       ComputeDataType,
                                                                        YDataType,
                                                                        PassThrough,
                                                                        Rank,
@@ -109,7 +109,7 @@ bool profile_layernorm_impl(int do_verification,
                                                                                  GammaDataType,
                                                                                  BetaDataType,
                                                                                  YDataType,
-                                                                                 AccDataType,
+                                                                                 ComputeDataType,
                                                                                  PassThrough,
                                                                                  Rank,
                                                                                  NumReduceDim>;
@@ -181,8 +181,8 @@ bool profile_layernorm_impl(int do_verification,
         {
             y_dev.FromDevice(y.mData.data());
 
-            bool pass = ck::utils::check_err(
-                y.mData, host_y.mData, "Error: Incorrect results d1", 1e-3, 1e-3);
+            bool pass =
+                ck::utils::check_err(y.mData, host_y.mData, "Error: Incorrect results", 1e-3, 1e-3);
 
             if(do_log)
             {
diff --git a/profiler/include/profiler/profile_pool2d_fwd_impl.hpp b/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..0c888db1f47a6511de4822c1b27a23b2a2d6e814
--- /dev/null
+++ b/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
@@ -0,0 +1,264 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+bool profile_pool2d_fwd_impl(int do_verification,
+                             int init_method,
+                             bool do_log,
+                             bool time_kernel,
+                             std::vector<index_t> in_length, // NCHW
+                             std::vector<index_t> window_spatial_lengths,
+                             std::vector<index_t> window_strides,
+                             std::vector<index_t> input_left_pads,
+                             std::vector<index_t> input_right_pads)
+{
+    constexpr index_t InOutRank  = 4;
+    constexpr index_t WindowRank = 2;
+
+    if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
+       window_strides.size() != WindowRank || input_left_pads.size() != WindowRank ||
+       input_right_pads.size() != WindowRank)
+        return false;
+
+    std::vector<index_t> out_length(InOutRank);
+
+    int N = in_length[0];
+    int C = in_length[1];
+
+    out_length[0] = N;
+    out_length[1] = C;
+
+    // Calculate Ho, Wo
+    for(int i = 2; i < InOutRank; ++i)
+    {
+        auto pad1           = input_left_pads[i - 2];
+        auto pad2           = input_right_pads[i - 2];
+        auto windows_size   = window_spatial_lengths[i - 2];
+        auto windows_stride = window_strides[i - 2];
+        out_length[i]       = (in_length[i] + pad1 + pad2 - windows_size) / windows_stride + 1;
+    }
+
+    int Hi = in_length[2];
+    int Wi = in_length[3];
+    int Ho = out_length[2];
+    int Wo = out_length[3];
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+            return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
+        };
+
+    Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi));
+    Tensor<OutDataType> out_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    Tensor<OutDataType> out_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    switch(init_method)
+    {
+    case 0: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_1<InDataType>{}); break;
+    case 1: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5}); break;
+    default: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
+                                     out_indices_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
+
+    // add device normalization instances
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto instance_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
+
+    std::string best_instance_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    if(do_verification)
+    {
+        using ReferenceInstance = ck::tensor_operation::host::ReferencePoolingFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  ComputeDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  PropagateNan,
+                                                                                  OutputIndex>;
+
+        ReferenceInstance ref;
+        auto ref_argument = ref.MakeArgument(in_n_c_hi_wi,
+                                             out_n_c_ho_wo_host,
+                                             out_indices_n_c_ho_wo_host,
+                                             window_spatial_lengths,
+                                             window_strides,
+                                             input_left_pads,
+                                             input_right_pads);
+        auto ref_invoker  = ref.MakeInvoker();
+        ref_invoker.Run(ref_argument);
+    }
+
+    int num_kernel = 0;
+
+    for(auto& inst_ptr : instance_ptrs)
+    {
+        auto argument_ptr = inst_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            {C * Hi * Wi, 1, Wi * C, C},
+            {C * Ho * Wo, 1, Wo * C, C},
+            {C * Ho * Wo, 1, Wo * C, C},
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3});
+
+        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            ++num_kernel;
+        }
+        else
+        {
+            if(time_kernel)
+            {
+                std::cout << inst_ptr->GetTypeString() << " skipped due to unsupported argument: ";
+                LogRange(std::cout << "input lengths = ", in_length, ", ") << std::endl;
+            }
+
+            continue;
+        }
+
+        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
+
+        float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        std::size_t num_bytes = in_n_c_hi_wi.mDesc.GetElementSize() * sizeof(InDataType) +
+                                out_n_c_ho_wo_host.mDesc.GetElementSize() * sizeof(OutDataType);
+
+        if constexpr(OutputIndex)
+            num_bytes += out_indices_n_c_ho_wo_host.mDesc.GetElementSize() * sizeof(IndexDataType);
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        if(time_kernel)
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << inst_ptr->GetTypeString() << std::endl;
+
+        if(avg_time < best_avg_time)
+        {
+            best_instance_name = inst_ptr->GetTypeString();
+            best_avg_time      = avg_time;
+            best_gb_per_sec    = gb_per_sec;
+        }
+
+        if(do_verification)
+        {
+            out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
+
+            bool pass = ck::utils::check_err(out_n_c_ho_wo_device.mData,
+                                             out_n_c_ho_wo_host.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+
+            if constexpr(OutputIndex)
+            {
+                out_indices_device_buf.FromDevice(out_indices_n_c_ho_wo_device.mData.data());
+
+                pass = pass && ck::utils::check_err(out_indices_n_c_ho_wo_device,
+                                                    out_indices_n_c_ho_wo_host);
+            }
+
+            if(do_log)
+            {
+                LogRangeAsType<float>(std::cout << "in_n_c_hi_wi  : ", in_n_c_hi_wi.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_ho_wo_host  : ", out_n_c_ho_wo_host.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_ho_wo_device  : ", out_n_c_ho_wo_device.mData, ",")
+                    << std::endl;
+
+                if constexpr(OutputIndex)
+                    LogRangeAsType<float>(std::cout << "out_indices_n_c_ho_wo_device  : ",
+                                          out_indices_n_c_ho_wo_device.mData,
+                                          ",")
+                        << std::endl;
+            }
+
+            if(!pass)
+            {
+                std::cout << inst_ptr->GetTypeString() << " failed verification: ";
+                LogRange(std::cout << "lengths = [", in_length, ", ") << "]." << std::endl;
+                return false;
+            }
+            else
+            {
+                if(time_kernel)
+                    std::cout << "pass" << std::endl;
+            }
+        }
+    }
+
+    if(time_kernel)
+    {
+        LogRange(std::cout << "length = ", in_length, ",") << std::endl;
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
+    }
+
+    return true;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_pool3d_fwd_impl.hpp b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..41b57fd85332b6007e3c4d44ef11574ddf044498
--- /dev/null
+++ b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
@@ -0,0 +1,271 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+bool profile_pool3d_fwd_impl(int do_verification,
+                             int init_method,
+                             bool do_log,
+                             bool time_kernel,
+                             std::vector<index_t> in_length, // NCDHW
+                             std::vector<index_t> window_spatial_lengths,
+                             std::vector<index_t> window_strides,
+                             std::vector<index_t> input_left_pads,
+                             std::vector<index_t> input_right_pads)
+{
+    constexpr index_t InOutRank  = 5;
+    constexpr index_t WindowRank = 3;
+
+    if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
+       window_strides.size() != WindowRank || input_left_pads.size() != WindowRank ||
+       input_right_pads.size() != WindowRank)
+        return false;
+
+    std::vector<index_t> out_length(InOutRank);
+
+    int N = in_length[0];
+    int C = in_length[1];
+
+    out_length[0] = N;
+    out_length[1] = C;
+
+    // Calculate Do, Ho, Wo
+    for(int i = 2; i < InOutRank; ++i)
+    {
+        auto pad1           = input_left_pads[i - 2];
+        auto pad2           = input_right_pads[i - 2];
+        auto windows_size   = window_spatial_lengths[i - 2];
+        auto windows_stride = window_strides[i - 2];
+        out_length[i]       = (in_length[i] + pad1 + pad2 - windows_size) / windows_stride + 1;
+    }
+
+    int Di = in_length[2];
+    int Hi = in_length[3];
+    int Wi = in_length[4];
+    int Do = out_length[2];
+    int Ho = out_length[3];
+    int Wo = out_length[4];
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t D, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+        };
+
+    Tensor<InDataType> in_n_c_di_hi_wi(f_host_tensor_descriptor(N, C, Di, Hi, Wi));
+    Tensor<OutDataType> out_n_c_do_ho_wo_host(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_host(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+
+    Tensor<OutDataType> out_n_c_do_ho_wo_device(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_device(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+
+    switch(init_method)
+    {
+    case 0: in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_1<InDataType>{}); break;
+    case 1: in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5}); break;
+    default: in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_di_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
+                                     out_indices_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_di_hi_wi.mData.data());
+
+    // add device normalization instances
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto instance_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
+
+    std::string best_instance_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    if(do_verification)
+    {
+        using ReferenceInstance = ck::tensor_operation::host::ReferencePoolingFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  ComputeDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  PropagateNan,
+                                                                                  OutputIndex>;
+
+        ReferenceInstance ref;
+        auto ref_argument = ref.MakeArgument(in_n_c_di_hi_wi,
+                                             out_n_c_do_ho_wo_host,
+                                             out_indices_n_c_do_ho_wo_host,
+                                             window_spatial_lengths,
+                                             window_strides,
+                                             input_left_pads,
+                                             input_right_pads);
+        auto ref_invoker  = ref.MakeInvoker();
+        ref_invoker.Run(ref_argument);
+    }
+
+    int num_kernel = 0;
+
+    for(auto& inst_ptr : instance_ptrs)
+    {
+        auto argument_ptr = inst_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C},
+            {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+            {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3, 4});
+
+        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            ++num_kernel;
+        }
+        else
+        {
+            if(time_kernel)
+            {
+                std::cout << inst_ptr->GetTypeString() << " skipped due to unsupported argument: ";
+                LogRange(std::cout << "input lengths = ", in_length, ", ") << std::endl;
+            }
+
+            continue;
+        }
+
+        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
+
+        float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        std::size_t num_bytes = in_n_c_di_hi_wi.mDesc.GetElementSize() * sizeof(InDataType) +
+                                out_n_c_do_ho_wo_host.mDesc.GetElementSize() * sizeof(OutDataType);
+
+        if constexpr(OutputIndex)
+            num_bytes +=
+                out_indices_n_c_do_ho_wo_host.mDesc.GetElementSize() * sizeof(IndexDataType);
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        if(time_kernel)
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << inst_ptr->GetTypeString() << std::endl;
+
+        if(avg_time < best_avg_time)
+        {
+            best_instance_name = inst_ptr->GetTypeString();
+            best_avg_time      = avg_time;
+            best_gb_per_sec    = gb_per_sec;
+        }
+
+        if(do_verification)
+        {
+            out_device_buf.FromDevice(out_n_c_do_ho_wo_device.mData.data());
+
+            bool pass = ck::utils::check_err(out_n_c_do_ho_wo_device.mData,
+                                             out_n_c_do_ho_wo_host.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+
+            if constexpr(OutputIndex)
+            {
+                out_indices_device_buf.FromDevice(out_indices_n_c_do_ho_wo_device.mData.data());
+
+                pass = pass && ck::utils::check_err(out_indices_n_c_do_ho_wo_device,
+                                                    out_indices_n_c_do_ho_wo_host);
+            }
+
+            if(do_log)
+            {
+                LogRangeAsType<float>(
+                    std::cout << "in_n_c_di_hi_wi  : ", in_n_c_di_hi_wi.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_do_ho_wo_host  : ", out_n_c_do_ho_wo_host.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_do_ho_wo_device  : ", out_n_c_do_ho_wo_device.mData, ",")
+                    << std::endl;
+
+                if constexpr(OutputIndex)
+                    LogRangeAsType<float>(std::cout << "out_indices_n_c_do_ho_wo_device  : ",
+                                          out_indices_n_c_do_ho_wo_device.mData,
+                                          ",")
+                        << std::endl;
+            }
+
+            if(!pass)
+            {
+                std::cout << inst_ptr->GetTypeString() << " failed verification: ";
+                LogRange(std::cout << "lengths = [", in_length, ", ") << "]." << std::endl;
+                return false;
+            }
+            else
+            {
+                if(time_kernel)
+                    std::cout << "pass" << std::endl;
+            }
+        }
+    }
+
+    if(time_kernel)
+    {
+        LogRange(std::cout << "length = ", in_length, ",") << std::endl;
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
+    }
+
+    return true;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_reduce_impl.hpp b/profiler/include/profiler/profile_reduce_impl.hpp
index ccb99398f22fc9db702a6801a0858d9456c6b855..b54aa65aef727b048458e0b63f638bad143e4e28 100644
--- a/profiler/include/profiler/profile_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_reduce_impl.hpp
@@ -1,16 +1,16 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/device_reduce.hpp"
 
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp"
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_reduction.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_reduce.hpp"
 #include "ck/library/utility/host_common_util.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 
@@ -158,11 +158,6 @@ bool profile_reduce_impl_impl(bool do_verification,
 
     constexpr bool OutputIndex = (op_support_indices && UseIndex);
 
-    constexpr bool out_support_atomic_add = std::is_same<OutDataType, float>::value;
-    constexpr bool op_support_atomic_add =
-        !op_support_indices && ReduceOpId != ReduceTensorOp::NORM2;
-    constexpr bool use_atomic_add = (out_support_atomic_add && op_support_atomic_add);
-
     // 1) If InDataType is half_t, must use half_t as AccDataType for indexable reduction operations
     // 2) If InDataType is half_t, must use float as AccDataType for non-indexable reduction
     // operations
@@ -200,7 +195,8 @@ bool profile_reduce_impl_impl(bool do_verification,
     constexpr bool invalid_reduce = (invalid_reduce_1 || invalid_reduce_2 || invalid_reduce_3 ||
                                      invalid_reduce_4 || invalid_reduce_5 || invalid_reduce_6);
 
-    bool pass = true;
+    int num_kernel = 0;
+    bool pass      = true;
 
     if constexpr(!invalid_reduce)
     {
@@ -286,75 +282,25 @@ bool profile_reduce_impl_impl(bool do_verification,
             reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
                 static_cast<int32_t>(reduce_total_length));
 
-        using DeviceReduceInstPtr =
-            DeviceReducePtr<Rank, NumReduceDim, InElementwiseOperation, AccElementwiseOperation>;
-
-        std::vector<DeviceReduceInstPtr> reduce_ptrs;
-
-        add_device_reduce_instance_threadwise<InDataType,
-                                              AccDataType,
-                                              OutDataType,
-                                              Rank,
-                                              NumReduceDim,
-                                              ReduceOperation,
-                                              InElementwiseOperation,
-                                              AccElementwiseOperation,
-                                              PropagateNan,
-                                              UseIndex>(reduce_ptrs);
-
-        add_device_reduce_instance_blockwise<InDataType,
-                                             AccDataType,
-                                             OutDataType,
-                                             Rank,
-                                             NumReduceDim,
-                                             ReduceOperation,
-                                             InElementwiseOperation,
-                                             AccElementwiseOperation,
-                                             PropagateNan,
-                                             UseIndex>(reduce_ptrs);
-
-        if constexpr(use_atomic_add)
-        {
-            add_device_reduce_instance_multiblock_atomic_add<InDataType,
-                                                             AccDataType,
-                                                             OutDataType,
-                                                             Rank,
-                                                             NumReduceDim,
-                                                             ReduceOperation,
-                                                             InElementwiseOperation,
-                                                             AccElementwiseOperation,
-                                                             PropagateNan,
-                                                             UseIndex>(reduce_ptrs);
-        }
+        using ReduceOp = ck::tensor_operation::device::DeviceReduce<InDataType,
+                                                                    AccDataType,
+                                                                    OutDataType,
+                                                                    Rank,
+                                                                    NumReduceDim,
+                                                                    ReduceOperation,
+                                                                    InElementwiseOperation,
+                                                                    AccElementwiseOperation,
+                                                                    PropagateNan,
+                                                                    OutputIndex>;
+        const auto reduce_ptrs =
+            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+                ReduceOp>::GetInstances();
 
         if(reduce_ptrs.empty())
         {
             throw std::runtime_error("Wrong! No device REDUCE instance found");
         };
 
-        if(do_verification)
-        {
-            ReductionHost<InDataType,
-                          AccDataType,
-                          OutDataType,
-                          ReduceOperation,
-                          InElementwiseOperation,
-                          AccElementwiseOperation,
-                          Rank,
-                          NumReduceDim,
-                          PropagateNan,
-                          OutputIndex>
-                hostReduce(in.mDesc, out_ref.mDesc, invariantDims, reduceDims);
-
-            hostReduce.Run(alpha,
-                           in.mData.data(),
-                           beta,
-                           out_ref.mData.data(),
-                           out_indices_ref.mData.data(),
-                           in_elementwise_op,
-                           acc_elementwise_op);
-        };
-
         std::array<index_t, Rank> arrInLengths;
         std::array<index_t, Rank> arrInStrides;
         std::array<index_t, NumOutDim> arrOutLengths;
@@ -365,6 +311,49 @@ bool profile_reduce_impl_impl(bool do_verification,
         ck::ranges::copy(outLengths, arrOutLengths.begin());
         ck::ranges::copy(outStrides, arrOutStrides.begin());
 
+        if(do_verification)
+        {
+            using ReferenceReduceInstance =
+                ck::tensor_operation::host::ReferenceReduce<InDataType,
+                                                            AccDataType,
+                                                            OutDataType,
+                                                            Rank,
+                                                            NumReduceDim,
+                                                            ReduceOperation,
+                                                            InElementwiseOperation,
+                                                            AccElementwiseOperation,
+                                                            PropagateNan,
+                                                            OutputIndex>;
+
+            auto reduce_ref = ReferenceReduceInstance{};
+
+            auto argument_ptr_ref = reduce_ref.MakeArgumentPointer(arrInLengths,
+                                                                   arrInStrides,
+                                                                   arrOutLengths,
+                                                                   arrOutStrides,
+                                                                   reduceDims,
+                                                                   static_cast<double>(alpha),
+                                                                   static_cast<double>(beta),
+                                                                   in.mData.data(),
+                                                                   nullptr,
+                                                                   out_ref.mData.data(),
+                                                                   out_indices_ref.mData.data(),
+                                                                   in_elementwise_op,
+                                                                   acc_elementwise_op);
+
+            if(!reduce_ref.IsSupportedArgument(argument_ptr_ref.get()))
+            {
+                std::cout
+                    << "The runtime parameters not supported by the reduce reference, exiting!"
+                    << std::endl;
+                return (false);
+            };
+
+            auto invoker_ptr_ref = reduce_ref.MakeInvokerPointer();
+
+            (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
+        };
+
         for(auto& reduce_ptr : reduce_ptrs)
         {
             auto argument_ptr = reduce_ptr->MakeArgumentPointer(arrInLengths,
@@ -372,8 +361,8 @@ bool profile_reduce_impl_impl(bool do_verification,
                                                                 arrOutLengths,
                                                                 arrOutStrides,
                                                                 reduceDims,
-                                                                alpha,
-                                                                beta,
+                                                                static_cast<double>(alpha),
+                                                                static_cast<double>(beta),
                                                                 in_dev.GetDeviceBuffer(),
                                                                 nullptr,
                                                                 out_dev.GetDeviceBuffer(),
@@ -383,6 +372,8 @@ bool profile_reduce_impl_impl(bool do_verification,
 
             if(!reduce_ptr->IsSupportedArgument(argument_ptr.get()))
                 continue;
+            else
+                num_kernel++;
 
             std::string reduce_name = reduce_ptr->GetTypeString();
 
@@ -446,14 +437,20 @@ bool profile_reduce_impl_impl(bool do_verification,
             };
         };
 
-        if(time_kernel)
+        if(time_kernel && num_kernel > 0)
             std::cout << "Best Perf: " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s"
                       << std::endl;
     }
     else
     {
-        std::cout << "The requested reduction operation is not supported, please check !!!"
-                  << std::endl;
+        throw std::runtime_error(
+            "The requested reduction operation is not supported, please check!");
+    };
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
     };
 
     return pass;
diff --git a/profiler/include/profiler/profile_softmax_impl.hpp b/profiler/include/profiler/profile_softmax_impl.hpp
index 090cdaaa9a2a8b68701ea3e9adb5d8c678a5355d..daaf565149784729fc66c4c6ef048df47c0c8dae 100644
--- a/profiler/include/profiler/profile_softmax_impl.hpp
+++ b/profiler/include/profiler/profile_softmax_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -40,7 +40,11 @@ template <> std::string type_to_string<int8_t>()  { return "int8"; }
 template <> std::string type_to_string<int32_t>() { return "int32"; }
 // clang-format on
 
-template <typename InDataType, typename AccDataType, typename OutDataType, index_t Rank>
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
+          index_t NumReduceDim>
 bool profile_softmax_impl(int do_verification,
                           int init_method,
                           bool do_log,
@@ -48,13 +52,19 @@ bool profile_softmax_impl(int do_verification,
                           std::vector<index_t> in_length,
                           std::vector<index_t> in_strides,
                           std::vector<index_t> reduce_dims,
-                          AccDataType alpha,
-                          AccDataType beta)
+                          double alpha,
+                          double beta)
 {
     if(Rank != in_length.size())
     {
         throw std::runtime_error("Input tensor rank is different from template argument Rank!");
-    }
+    };
+
+    if(NumReduceDim != reduce_dims.size())
+    {
+        throw std::runtime_error(
+            "Input reduce_dims rank is different from template argument NumReduceDim!");
+    };
 
     Tensor<InDataType> in = in_strides.empty() ? Tensor<InDataType>(in_length)
                                                : Tensor<InDataType>(in_length, in_strides);
@@ -92,8 +102,13 @@ bool profile_softmax_impl(int do_verification,
 
     // add device softmax instances
     using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-    using DeviceOp    = tensor_operation::device::
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>;
+    using DeviceOp    = tensor_operation::device::DeviceSoftmax<InDataType,
+                                                             AccDataType,
+                                                             OutDataType,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             Rank,
+                                                             NumReduceDim>;
 
     // get device op instances
     const auto instances = tensor_operation::device::instance::DeviceOperationInstanceFactory<
@@ -112,18 +127,11 @@ bool profile_softmax_impl(int do_verification,
 
     for(auto& inst_ptr : instances)
     {
-        // Is this user's responsibility to check if problem mismatches kernel instance (ie. rank 3
-        // problem to rank 4 kernel) other than invoking IsSupportedArgument()?
-        if(!(inst_ptr->GetNumReduceDim() == static_cast<index_t>(reduce_dims.size())))
-        {
-            continue;
-        }
-
         auto argument_ptr = inst_ptr->MakeArgumentPointer(in_tensor_lengths,
                                                           in_tensor_strides,
                                                           reduce_dims,
-                                                          &alpha,
-                                                          &beta,
+                                                          alpha,
+                                                          beta,
                                                           in_dev.GetDeviceBuffer(),
                                                           out_dev.GetDeviceBuffer(),
                                                           PassThrough{},
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index bc87554cecec617659b9b05200a7e3dcdfbb3fd7..3f19096f48db259239de51a5dfc82059b0aca73a 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -6,7 +6,9 @@ set(PROFILER_SOURCES
     profile_gemm_bilinear.cpp
     profile_gemm_bias_add_reduce.cpp
     profile_gemm_add_add_fastgelu.cpp
+    profile_gemm_add_multiply.cpp
     profile_gemm_add_fastgelu.cpp
+    profile_gemm_add_relu_add_layernorm.cpp
     profile_gemm_fastgelu.cpp
     profile_gemm_reduce.cpp
     profile_batched_gemm.cpp
@@ -23,9 +25,17 @@ set(PROFILER_SOURCES
     profile_reduce.cpp
     profile_groupnorm.cpp
     profile_layernorm.cpp
+    profile_avg_pool2d_fwd.cpp
+    profile_max_pool3d_fwd.cpp
     profile_softmax.cpp
     profile_batchnorm_fwd.cpp
     profile_batchnorm_bwd.cpp
+    profile_batchnorm_infer.cpp
+    profile_grouped_gemm_fastgelu.cpp
+    profile_contraction_bilinear.cpp
+    profile_contraction_scale.cpp
+    profile_batched_gemm_multi_d.cpp
+    profile_grouped_conv_bwd_data.cpp
 )
 
 set(PROFILER_EXECUTABLE ckProfiler)
@@ -38,8 +48,10 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
@@ -63,5 +75,10 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_instan
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
-
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/profiler/src/profile_avg_pool2d_fwd.cpp b/profiler/src/profile_avg_pool2d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c67897c04421085b69cb03e3ce97faf259d4f80b
--- /dev/null
+++ b/profiler/src/profile_avg_pool2d_fwd.cpp
@@ -0,0 +1,141 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+#include <unordered_map>
+
+#include "profiler/data_type_enum.hpp"
+#include "profiler/profile_pool2d_fwd_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+using ck::index_t;
+
+struct avgPoolFwdArgParser
+{
+    std::unordered_map<std::string, std::vector<int>> long_opts = {
+        {"length", {}}, {"wsize", {}}, {"wstride", {}}, {"pad1", {}}, {"pad2", {}}};
+
+    bool parse_opt(int argc, char* argv[], const std::string& key, int i)
+    {
+        if(std::string("--") + key == argv[i])
+        {
+            int pos = i;
+            while(++i < argc && argv[i][0] != '-') {}
+            int end = i;
+            for(int j = pos + 1; j < end; j++)
+            {
+                long_opts[key].push_back(std::stoi(argv[j]));
+            }
+            return true;
+        }
+        return false;
+    }
+
+    void operator()(int argc, char* argv[])
+    {
+        for(auto& kv : long_opts)
+        {
+            for(int i = 1; i < argc; i++)
+            {
+                if(parse_opt(argc, argv, kv.first, i))
+                    break;
+            }
+        }
+    }
+};
+
+void print_help_avg_pool2d_fwd()
+{
+    std::cout << "arg1: data type (0: fp16; 1: fp32)\n"
+              << "arg2: verification (0: no; 1: yes)\n"
+              << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+              << "arg4: print tensor value (0: no; 1: yes)\n"
+              << "arg5: time kernel (0=no, 1=yes)\n"
+              << "--length: input tensor length for NDHW(e.g, --length 2 32 30 30) \n"
+              << "--wsize: window size for YX (e.g, --wsize 2 2) \n"
+              << "--wstride: window stride for HW (e.g, --wstride 2 2) \n"
+              << "--pad1: left side of padding in HW (e.g, --pad1 1 1) \n"
+              << "--pad2: right side of padding in HW (e.g, --pad2 1 1) \n"
+              << "eg: ckProfiler avg_pool2d_fwd 0 1 2 0 1 0 --length 2 32 30 30 --wsize 2 2 "
+                 "--wstride 2 2 --pad1 1 1 --pad2 1 1"
+              << std::endl;
+}
+
+int profile_avg_pool2d_fwd(int argc, char* argv[])
+{
+    ck::DataTypeEnum data_type = ck::DataTypeEnum::Half;
+    bool do_verification       = true;
+    int init_method            = 0;
+    bool do_log                = false;
+    bool time_kernel           = true;
+
+    std::vector<index_t> in_length = {2, 32, 30, 30};
+    std::vector<index_t> wsize     = {2, 2};
+    std::vector<index_t> wstride   = {2, 2};
+    std::vector<index_t> pad1      = {1, 1};
+    std::vector<index_t> pad2      = {1, 1};
+
+    if(argc != 2 && argc != 25)
+    {
+        print_help_avg_pool2d_fwd();
+        return 0;
+    }
+    else if(argc == 25)
+    {
+        data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
+        do_verification = std::stoi(argv[3]);
+        init_method     = std::stoi(argv[4]);
+        do_log          = std::stoi(argv[5]);
+        time_kernel     = std::stoi(argv[6]);
+
+        // parse the long options
+        avgPoolFwdArgParser arg_parser;
+        arg_parser(argc, argv);
+        in_length = arg_parser.long_opts["length"];
+        wsize     = arg_parser.long_opts["wsize"];
+        wstride   = arg_parser.long_opts["wstride"];
+        pad1      = arg_parser.long_opts["pad1"];
+        pad2      = arg_parser.long_opts["pad2"];
+    }
+
+    using F16                 = ck::half_t;
+    using F32                 = float;
+    using I32                 = int32_t;
+    constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+    if(data_type == ck::DataTypeEnum::Half)
+    {
+        ck::profiler::profile_pool2d_fwd_impl<F16, F16, F32, I32, ReduceOpId, false, false>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            in_length,
+            wsize,
+            wstride,
+            pad1,
+            pad2);
+    }
+    else if(data_type == ck::DataTypeEnum::Float)
+    {
+        ck::profiler::profile_pool2d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, false>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            in_length,
+            wsize,
+            wstride,
+            pad1,
+            pad2);
+    }
+    else
+    {
+        throw std::runtime_error("not implemented yet");
+    }
+
+    return 0;
+}
+
+REGISTER_PROFILER_OPERATION("avg_pool2d_fwd", "avg_pool2d fwd", profile_avg_pool2d_fwd);
diff --git a/profiler/src/profile_batched_gemm.cpp b/profiler/src/profile_batched_gemm.cpp
index 907a373794f2d8a59a066b94a591c4890c5c4038..222532b7bbdeb36bfa9953fb9f174baef3b548ac 100644
--- a/profiler/src/profile_batched_gemm.cpp
+++ b/profiler/src/profile_batched_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdint>
 #include <iostream>
@@ -10,6 +10,8 @@
 #include "profiler/profile_batched_gemm_impl.hpp"
 #include "profiler_operation_registry.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 enum struct GemmMatrixLayout
 {
     MK_KN_MN, // 0
@@ -78,55 +80,72 @@ int profile_batched_gemm(int argc, char* argv[])
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
 
-    auto profile = [&](auto a_type,
-                       auto b_type,
-                       auto c_type,
-                       auto a_layout,
-                       auto b_layout,
-                       auto c_layout) {
-        using ADataType = decltype(a_type);
-        using BDataType = decltype(b_type);
-        using CDataType = decltype(c_type);
-
-        using ALayout = decltype(a_layout);
-        using BLayout = decltype(b_layout);
-        using CLayout = decltype(c_layout);
-
-        const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
-        const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
-        const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
-
-        const int StrideA_ = (StrideA < 0) ? DefaultStrideA : StrideA;
-        const int StrideB_ = (StrideB < 0) ? DefaultStrideB : StrideB;
-        const int StrideC_ = (StrideC < 0) ? DefaultStrideC : StrideC;
-
-        const int DefaultBatchStrideA = (ck::is_same_v<ALayout, Row> ? M : K) * StrideA_;
-        const int DefaultBatchStrideB = (ck::is_same_v<BLayout, Row> ? K : N) * StrideB_;
-        const int DefaultBatchStrideC = (ck::is_same_v<CLayout, Row> ? M : N) * StrideC_;
-
-        const int BatchStrideA_ = (BatchStrideA < 0) ? DefaultBatchStrideA : BatchStrideA;
-        const int BatchStrideB_ = (BatchStrideB < 0) ? DefaultBatchStrideB : BatchStrideB;
-        const int BatchStrideC_ = (BatchStrideC < 0) ? DefaultBatchStrideC : BatchStrideC;
-
-        bool pass = ck::profiler::
-            profile_batched_gemm_impl<ADataType, BDataType, CDataType, ALayout, BLayout, CLayout>(
-                do_verification,
-                init_method,
-                do_log,
-                time_kernel,
-                M,
-                N,
-                K,
-                BatchStrideA_,
-                BatchStrideB_,
-                BatchStrideC_,
-                StrideA_,
-                StrideB_,
-                StrideC_,
-                BatchCount);
-
-        return pass ? 0 : 1;
-    };
+    auto profile =
+        [&](auto a_type, auto b_type, auto c_type, auto a_layout, auto b_layout, auto c_layout) {
+            using ADataType = decltype(a_type);
+            using BDataType = decltype(b_type);
+            using CDataType = decltype(c_type);
+
+            using ALayout = decltype(a_layout);
+            using BLayout = decltype(b_layout);
+            using CLayout = decltype(c_layout);
+
+            const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
+            const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
+            const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
+
+            const int StrideA_ = (StrideA < 0) ? DefaultStrideA : StrideA;
+            const int StrideB_ = (StrideB < 0) ? DefaultStrideB : StrideB;
+            const int StrideC_ = (StrideC < 0) ? DefaultStrideC : StrideC;
+
+            const int DefaultBatchStrideA = (ck::is_same_v<ALayout, Row> ? M : K) * StrideA_;
+            const int DefaultBatchStrideB = (ck::is_same_v<BLayout, Row> ? K : N) * StrideB_;
+            const int DefaultBatchStrideC = (ck::is_same_v<CLayout, Row> ? M : N) * StrideC_;
+
+            const int BatchStrideA_ = (BatchStrideA < 0) ? DefaultBatchStrideA : BatchStrideA;
+            const int BatchStrideB_ = (BatchStrideB < 0) ? DefaultBatchStrideB : BatchStrideB;
+            const int BatchStrideC_ = (BatchStrideC < 0) ? DefaultBatchStrideC : BatchStrideC;
+
+            using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using CElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+            using DeviceOp = ck::tensor_operation::device::DeviceBatchedGemm<ALayout,
+                                                                             BLayout,
+                                                                             CLayout,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             AElementOp,
+                                                                             BElementOp,
+                                                                             CElementOp>;
+
+            bool pass = ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                                BDataType,
+                                                                CDataType,
+                                                                ALayout,
+                                                                BLayout,
+                                                                CLayout,
+                                                                AElementOp,
+                                                                BElementOp,
+                                                                CElementOp,
+                                                                DeviceOp>(do_verification,
+                                                                          init_method,
+                                                                          do_log,
+                                                                          time_kernel,
+                                                                          M,
+                                                                          N,
+                                                                          K,
+                                                                          BatchStrideA_,
+                                                                          BatchStrideB_,
+                                                                          BatchStrideC_,
+                                                                          StrideA_,
+                                                                          StrideB_,
+                                                                          StrideC_,
+                                                                          BatchCount);
+
+            return pass ? 0 : 1;
+        };
 
     if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
     {
diff --git a/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp b/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp
index f440a3094eb16b4eef1c5dee46cb08f6ab25d933..3d29c4b84aa5abfafbb220c46205413adb534770 100644
--- a/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp
+++ b/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_batched_gemm_gemm.cpp b/profiler/src/profile_batched_gemm_gemm.cpp
index 6015c93be35268c1c87c129d0cc7d2c9bdf0584f..9a99874d1c7041f692133a6ef2e792bdf0dd8105 100644
--- a/profiler/src/profile_batched_gemm_gemm.cpp
+++ b/profiler/src/profile_batched_gemm_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_batched_gemm_multi_d.cpp b/profiler/src/profile_batched_gemm_multi_d.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..98b462d9503438818554a8327b2f09d12943f154
--- /dev/null
+++ b/profiler/src/profile_batched_gemm_multi_d.cpp
@@ -0,0 +1,190 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdint>
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_batched_gemm_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp"
+
+enum struct GemmMatrixLayout
+{
+    MK_KN_MN, // 0
+    MK_NK_MN, // 1
+    KM_KN_MN, // 2
+    KM_NK_MN, // 3
+};
+
+enum struct GemmDataType
+{
+    F16_F16_F16,    // 0
+    INT8_INT8_INT8, // 1
+};
+
+#define OP_NAME "batched_gemm_multi_d"
+#define OP_DESC "Batched GEMM multi D"
+
+int profile_batched_gemm_multi_d(int argc, char* argv[])
+{
+    if(argc != 18)
+    {
+        // clang-format off
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp16; 1: int8)\n");
+        printf("arg3: matrix layout (0: A[g, m, k] * B[g, k, n] = C[g, m, n];\n");
+        printf("                     1: A[g, m, k] * B[g, n, k] = C[g, m, n];\n");
+        printf("                     2: A[g, k, m] * B[g, k, n] = C[g, m, n];\n");
+        printf("                     3: A[g, k, m] * B[g, n, k] = C[g, m, n])\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=n0, 1=yes)\n");
+        printf("arg8 to 17: M, N, K, StrideA, StrideB, StrideC, BatchStrideA, BatchStrideB, BatchStrideC, BatchCount\n");
+        // clang-format on
+        exit(1);
+    }
+
+    const auto data_type       = static_cast<GemmDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<GemmMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+
+    const int M = std::stoi(argv[8]);
+    const int N = std::stoi(argv[9]);
+    const int K = std::stoi(argv[10]);
+
+    const int StrideA = std::stoi(argv[11]);
+    const int StrideB = std::stoi(argv[12]);
+    const int StrideC = std::stoi(argv[13]);
+
+    const int BatchStrideA = std::stoi(argv[14]);
+    const int BatchStrideB = std::stoi(argv[15]);
+    const int BatchStrideC = std::stoi(argv[16]);
+
+    const int BatchCount = std::stoi(argv[17]);
+
+    using F16  = ck::half_t;
+    using INT8 = int8_t;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    auto profile =
+        [&](auto a_type, auto b_type, auto c_type, auto a_layout, auto b_layout, auto c_layout) {
+            using ADataType  = decltype(a_type);
+            using BDataType  = decltype(b_type);
+            using CDataType  = decltype(c_type);
+            using DsDataType = ck::Tuple<>;
+
+            using ALayout  = decltype(a_layout);
+            using BLayout  = decltype(b_layout);
+            using CLayout  = decltype(c_layout);
+            using DsLayout = ck::Tuple<>;
+
+            const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
+            const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
+            const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
+
+            const int StrideA_ = (StrideA < 0) ? DefaultStrideA : StrideA;
+            const int StrideB_ = (StrideB < 0) ? DefaultStrideB : StrideB;
+            const int StrideC_ = (StrideC < 0) ? DefaultStrideC : StrideC;
+
+            const int DefaultBatchStrideA = (ck::is_same_v<ALayout, Row> ? M : K) * StrideA_;
+            const int DefaultBatchStrideB = (ck::is_same_v<BLayout, Row> ? K : N) * StrideB_;
+            const int DefaultBatchStrideC = (ck::is_same_v<CLayout, Row> ? M : N) * StrideC_;
+
+            const int BatchStrideA_ = (BatchStrideA < 0) ? DefaultBatchStrideA : BatchStrideA;
+            const int BatchStrideB_ = (BatchStrideB < 0) ? DefaultBatchStrideB : BatchStrideB;
+            const int BatchStrideC_ = (BatchStrideC < 0) ? DefaultBatchStrideC : BatchStrideC;
+
+            using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using CElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+            using DeviceOp = ck::tensor_operation::device::DeviceBatchedGemmMultiD<ALayout,
+                                                                                   BLayout,
+                                                                                   DsLayout,
+                                                                                   CLayout,
+                                                                                   ADataType,
+                                                                                   BDataType,
+                                                                                   DsDataType,
+                                                                                   CDataType,
+                                                                                   AElementOp,
+                                                                                   BElementOp,
+                                                                                   CElementOp>;
+
+            bool pass = ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                                BDataType,
+                                                                CDataType,
+                                                                ALayout,
+                                                                BLayout,
+                                                                CLayout,
+                                                                AElementOp,
+                                                                BElementOp,
+                                                                CElementOp,
+                                                                DeviceOp>(do_verification,
+                                                                          init_method,
+                                                                          do_log,
+                                                                          time_kernel,
+                                                                          M,
+                                                                          N,
+                                                                          K,
+                                                                          BatchStrideA_,
+                                                                          BatchStrideB_,
+                                                                          BatchStrideC_,
+                                                                          StrideA_,
+                                                                          StrideB_,
+                                                                          StrideC_,
+                                                                          BatchCount);
+
+            return pass ? 0 : 1;
+        };
+
+    if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Col{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Col{}, Col{}, Row{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_batched_gemm_multi_d);
diff --git a/profiler/src/profile_batched_gemm_reduce.cpp b/profiler/src/profile_batched_gemm_reduce.cpp
index 6b1dfc01427a1cd34eefc89584bb6f3ce8bf206f..9620d63cafbeedfb5398b553b5172418923bb8e0 100644
--- a/profiler/src/profile_batched_gemm_reduce.cpp
+++ b/profiler/src/profile_batched_gemm_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_batchnorm_bwd.cpp b/profiler/src/profile_batchnorm_bwd.cpp
index 44ce7350ff06eaad3eff7d93f60c022d1a4c2b60..1738d53dbe7c6d2581e15df8bae58f7c48733999 100644
--- a/profiler/src/profile_batchnorm_bwd.cpp
+++ b/profiler/src/profile_batchnorm_bwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_batchnorm_fwd.cpp b/profiler/src/profile_batchnorm_fwd.cpp
index 902a1fc423f98880b43e1551f3305854d1a0b9b8..507fb4b450ba2d7582346458e729d958908950c0 100644
--- a/profiler/src/profile_batchnorm_fwd.cpp
+++ b/profiler/src/profile_batchnorm_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
@@ -148,7 +148,7 @@ int profile_batchnorm_forward(int argc, char* argv[])
     {
         if(arg_parser.inLengths.size() == 4 && arg_parser.reduceDims.size() == 3)
         {
-            profile_batchnorm_forward_impl<F16, F16, F32, F16, F16, F16, 4, 3>(
+            profile_batchnorm_forward_impl<F16, F16, F32, F16, F16, F32, 4, 3>(
                 arg_parser.do_verification,
                 arg_parser.init_method,
                 arg_parser.do_dumpout,
diff --git a/profiler/src/profile_batchnorm_infer.cpp b/profiler/src/profile_batchnorm_infer.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..f1c19bc36e8aded070b3ba843d9fa7007f37a55e
--- /dev/null
+++ b/profiler/src/profile_batchnorm_infer.cpp
@@ -0,0 +1,202 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+#include <getopt.h>
+
+#include "ck/library/utility/host_common_util.hpp"
+#include "profiler/profile_batchnorm_infer_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+using ck::index_t;
+
+using namespace std;
+
+static const struct option long_options[] = {{"inOutLengths", required_argument, nullptr, 'D'},
+                                             {"reduceDims", required_argument, nullptr, 'R'},
+                                             {"dumpout", required_argument, nullptr, 'o'},
+                                             {"verify", required_argument, nullptr, 'v'},
+                                             {"help", no_argument, nullptr, '?'},
+                                             {nullptr, 0, nullptr, 0}};
+
+class BatchnormInferArgParser
+{
+    private:
+    int option_index = 0;
+
+    public:
+    std::vector<size_t> inLengths;
+    std::vector<int> reduceDims;
+
+    bool do_verification = false;
+    bool do_dumpout      = false;
+
+    bool updateMovingAverage;
+    bool saveMeanAndInvVariance;
+
+    int data_type    = 0;
+    int init_method  = 2;
+    bool time_kernel = false;
+
+    BatchnormInferArgParser()  = default;
+    ~BatchnormInferArgParser() = default;
+
+    void show_usage(const char* cmd)
+    {
+        // clang-format off
+        std::cout << "Usage of " << cmd << std::endl;
+        std::cout << "--inOutLengths or -D, comma separated list of input tensor dimension lengths, must have 4 integers for nhwc" << std::endl;
+        std::cout << "--reduceDims or -R, comma separated list of dimensions to reduce on" << std::endl;  
+        std::cout << "--verify or -v, 1/0 to indicate whether to verify the result by comparing with the host-based batch-normalization" << std::endl;
+        std::cout << "Arg1: data type (0: fp16, 1: fp32, 5: bp16, 6: fp64)" << std::endl;
+        std::cout << "Arg2: init method used for bnScale and bnBias (0=no init, 1=single integer value, 2=scope integer value, 3=decimal value)" << std::endl;
+        std::cout << "Arg3: time kernel (0=no, 1=yes)" << std::endl;
+        // clang-format on
+    };
+
+    int operator()(int argc, char* argv[])
+    {
+        using ck::host_common::getTypeValuesFromString;
+
+        int ch;
+
+        optind++; // to skip the module name
+
+        while(1)
+        {
+            ch = getopt_long(argc, argv, "D:R:v:o:", long_options, &option_index);
+            if(ch == -1)
+                break;
+            switch(ch)
+            {
+            case 'D':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                inLengths = getTypeValuesFromString<size_t>(optarg);
+                break;
+            case 'R':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                reduceDims = getTypeValuesFromString<int>(optarg);
+                break;
+            case 'v':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                do_verification = static_cast<bool>(std::atoi(optarg));
+                break;
+            case 'o':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                do_dumpout = static_cast<bool>(std::atoi(optarg));
+                break;
+            case '?':
+                if(std::string(long_options[option_index].name) == "help")
+                {
+                    show_usage(argv[0]);
+                    return -1;
+                };
+                break;
+
+            default:
+                show_usage(argv[0]);
+                std::cerr << "Invalid cmd-line options!" << std::endl;
+                return -1;
+            };
+        };
+
+        if(optind + 3 > argc)
+            throw std::runtime_error("Invalid cmd-line arguments, more argumetns are needed!");
+
+        data_type   = std::atoi(argv[optind++]);
+        init_method = std::atoi(argv[optind++]);
+        time_kernel = static_cast<bool>(std::atoi(argv[optind++]));
+
+        if(data_type != 0 && data_type != 1 && data_type != 5 && data_type != 6)
+            return -1;
+
+        return 0;
+    };
+}; // end of class AppArgs
+
+static const double epsilon = std::numeric_limits<float>::epsilon();
+
+int profile_batchnorm_infer(int argc, char* argv[])
+{
+    using ck::profiler::profile_batchnorm_infer_impl;
+
+    BatchnormInferArgParser arg_parser;
+
+    if(arg_parser(argc, argv) != 0)
+        return -1;
+
+    using F16  = ck::half_t;
+    using F32  = float;
+    using BF16 = ck::bhalf_t;
+    using F64  = double;
+
+    if(arg_parser.data_type == 0)
+    {
+        if(arg_parser.inLengths.size() == 4 && arg_parser.reduceDims.size() == 3)
+        {
+            profile_batchnorm_infer_impl<F16, F16, F32, F16, F16, F32, 4, 3>(
+                arg_parser.do_verification,
+                arg_parser.init_method,
+                arg_parser.do_dumpout,
+                arg_parser.time_kernel,
+                arg_parser.inLengths,
+                arg_parser.reduceDims,
+                epsilon);
+        };
+    }
+    else if(arg_parser.data_type == 1)
+    {
+        if(arg_parser.inLengths.size() == 4 && arg_parser.reduceDims.size() == 3)
+        {
+            profile_batchnorm_infer_impl<F32, F32, F32, F32, F32, F32, 4, 3>(
+                arg_parser.do_verification,
+                arg_parser.init_method,
+                arg_parser.do_dumpout,
+                arg_parser.time_kernel,
+                arg_parser.inLengths,
+                arg_parser.reduceDims,
+                epsilon);
+        };
+    }
+    else if(arg_parser.data_type == 5)
+    {
+        if(arg_parser.inLengths.size() == 4 && arg_parser.reduceDims.size() == 3)
+        {
+            profile_batchnorm_infer_impl<BF16, BF16, F32, BF16, BF16, F32, 4, 3>(
+                arg_parser.do_verification,
+                arg_parser.init_method,
+                arg_parser.do_dumpout,
+                arg_parser.time_kernel,
+                arg_parser.inLengths,
+                arg_parser.reduceDims,
+                epsilon);
+        };
+    }
+    else if(arg_parser.data_type == 6)
+    {
+        if(arg_parser.inLengths.size() == 4 && arg_parser.reduceDims.size() == 3)
+        {
+            profile_batchnorm_infer_impl<F64, F64, F64, F64, F64, F64, 4, 3>(
+                arg_parser.do_verification,
+                arg_parser.init_method,
+                arg_parser.do_dumpout,
+                arg_parser.time_kernel,
+                arg_parser.inLengths,
+                arg_parser.reduceDims,
+                epsilon);
+        };
+    }
+
+    return 0;
+}
+
+REGISTER_PROFILER_OPERATION("bnorm_infer", "Batchnorm inference", profile_batchnorm_infer);
diff --git a/profiler/src/profile_contraction_bilinear.cpp b/profiler/src/profile_contraction_bilinear.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..6ed184120477c50d0adcb97e997038ad39945233
--- /dev/null
+++ b/profiler/src/profile_contraction_bilinear.cpp
@@ -0,0 +1,165 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <vector>
+
+#include "profiler/profile_contraction_impl.hpp"
+#include "profiler/profile_contraction_utils.hpp"
+#include "profiler_operation_registry.hpp"
+
+#define OP_NAME "contraction_bilinear"
+#define OP_DESC "CONTRACTION+Bilinear"
+
+static void print_helper_msg()
+{
+    std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+              << "arg2: data type (0: fp32; 1: f64)\n"
+              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
+              << "arg4: verification (0: no; 1: yes)\n"
+              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
+              << "value)\n"
+              << "arg6: print tensor value (0: no; 1: yes)\n"
+              << "arg7: time kernel (0: no, 1: yes)\n"
+              << "arg8 and arg9: alpha and beta\n"
+              << "arg10 to 15: M0, M1, N0, N1, K0, K1\n"
+              << "arg16 to 31: Strides for A, B, D and E (skip for default)\n"
+              << std::endl;
+}
+
+int profile_contraction_bilinear(int argc, char* argv[])
+{
+    const bool default_strides = argc == 16;
+
+    if(argc != 32 && argc != 16)
+    {
+        print_helper_msg();
+        exit(1);
+    }
+
+    const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification    = std::stoi(argv[4]);
+    const ck::index_t init_method = std::stoi(argv[5]);
+    const bool do_log             = std::stoi(argv[6]);
+    const bool time_kernel        = std::stoi(argv[7]);
+    const float alpha             = std::stof(argv[8]);
+    const float beta              = std::stof(argv[9]);
+
+    std::vector<ck::index_t> M;
+    std::vector<ck::index_t> N;
+    std::vector<ck::index_t> K;
+    const ck::index_t dims_arg_num = 10;
+    collect_index_params(argv, M, dims_arg_num, 2);
+    collect_index_params(argv, N, dims_arg_num + 2, 2);
+    collect_index_params(argv, K, dims_arg_num + 4, 2);
+
+    std::vector<ck::index_t> StridesA;
+    std::vector<ck::index_t> StridesB;
+    std::vector<ck::index_t> StridesE;
+    std::vector<ck::index_t> StridesD;
+    if(!default_strides)
+    {
+        collect_index_params(argv, StridesA, dims_arg_num + 6, 4);
+        collect_index_params(argv, StridesB, dims_arg_num + 10, 4);
+        collect_index_params(argv, StridesE, dims_arg_num + 14, 4);
+        collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
+    }
+
+    using F32 = float;
+    using F64 = double;
+
+    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
+        using ALayout   = decltype(a_layout);
+        using BLayout   = decltype(b_layout);
+        using CDELayout = decltype(cde_layout);
+
+        using DataType = decltype(type);
+
+        if(default_strides)
+        {
+            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+        }
+        bool pass = ck::profiler::profile_contraction_impl<ALayout,
+                                                           BLayout,
+                                                           CDELayout,
+                                                           DataType,
+                                                           ck::Tuple<DataType>,
+                                                           Bilinear>(do_verification,
+                                                                     init_method,
+                                                                     do_log,
+                                                                     time_kernel,
+                                                                     Bilinear{alpha, beta},
+                                                                     M,
+                                                                     N,
+                                                                     K,
+                                                                     StridesA,
+                                                                     StridesB,
+                                                                     StridesE,
+                                                                     StridesD);
+
+        return pass;
+    };
+
+    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+       layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F64{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_bilinear);
diff --git a/profiler/src/profile_contraction_scale.cpp b/profiler/src/profile_contraction_scale.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..6784b916f608811baeb361c5ea303c3d2eb145ca
--- /dev/null
+++ b/profiler/src/profile_contraction_scale.cpp
@@ -0,0 +1,162 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <vector>
+
+#include "profiler/profile_contraction_impl.hpp"
+#include "profiler/profile_contraction_utils.hpp"
+#include "profiler_operation_registry.hpp"
+
+#define OP_NAME "contraction_scale"
+#define OP_DESC "CONTRACTION+Scale"
+
+static void print_helper_msg()
+{
+    std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+              << "arg2: data type (0: fp32; 1: f64)\n"
+              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
+              << "arg4: verification (0: no; 1: yes)\n"
+              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
+              << "value)\n"
+              << "arg6: print tensor value (0: no; 1: yes)\n"
+              << "arg7: time kernel (0: no, 1: yes)\n"
+              << "arg8: alpha\n"
+              << "arg9 to 14: M0, M1, N0, N1, K0, K1\n"
+              << "arg15 to 30: Strides for A, B, D and E (skip for default)\n"
+              << std::endl;
+}
+
+int profile_contraction_scale(int argc, char* argv[])
+{
+    const bool default_strides = argc == 15;
+
+    if(argc != 31 && argc != 15)
+    {
+        print_helper_msg();
+        exit(1);
+    }
+
+    const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification    = std::stoi(argv[4]);
+    const ck::index_t init_method = std::stoi(argv[5]);
+    const bool do_log             = std::stoi(argv[6]);
+    const bool time_kernel        = std::stoi(argv[7]);
+    const float alpha             = std::stof(argv[8]);
+
+    std::vector<ck::index_t> M;
+    std::vector<ck::index_t> N;
+    std::vector<ck::index_t> K;
+    const ck::index_t dims_arg_num = 9;
+    collect_index_params(argv, M, dims_arg_num, 2);
+    collect_index_params(argv, N, dims_arg_num + 2, 2);
+    collect_index_params(argv, K, dims_arg_num + 4, 2);
+
+    std::vector<ck::index_t> StridesA;
+    std::vector<ck::index_t> StridesB;
+    std::vector<ck::index_t> StridesE;
+    std::vector<ck::index_t> StridesD;
+    if(!default_strides)
+    {
+        collect_index_params(argv, StridesA, dims_arg_num + 6, 4);
+        collect_index_params(argv, StridesB, dims_arg_num + 10, 4);
+        collect_index_params(argv, StridesE, dims_arg_num + 14, 4);
+        collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
+    }
+
+    using F32 = float;
+    using F64 = double;
+
+    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
+        using ALayout   = decltype(a_layout);
+        using BLayout   = decltype(b_layout);
+        using CDELayout = decltype(cde_layout);
+
+        using DataType = decltype(type);
+
+        if(default_strides)
+        {
+            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+        }
+
+        bool pass = ck::profiler::
+            profile_contraction_impl<ALayout, BLayout, CDELayout, DataType, ck::Tuple<>, Scale>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                Scale{alpha},
+                M,
+                N,
+                K,
+                StridesA,
+                StridesB,
+                StridesE,
+                StridesD);
+
+        return pass;
+    };
+
+    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+       layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F64{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_scale);
diff --git a/profiler/src/profile_conv_bwd_data.cpp b/profiler/src/profile_conv_bwd_data.cpp
index 9241ead738ef8a8a67d8682d743f6c9bd0faf640..465abacc4a024f7a8ed8fc0ac5feb0c0e6b13fc0 100644
--- a/profiler/src/profile_conv_bwd_data.cpp
+++ b/profiler/src/profile_conv_bwd_data.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_conv_fwd.cpp b/profiler/src/profile_conv_fwd.cpp
index b57ee7fd94261f4437d1b62f0b6450ea74522db9..701999d8a9fcd3718fc8d62e9b77f4f49159d474 100644
--- a/profiler/src/profile_conv_fwd.cpp
+++ b/profiler/src/profile_conv_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_conv_fwd_bias_relu.cpp b/profiler/src/profile_conv_fwd_bias_relu.cpp
index b44007cde4742efcec188f0b620e2cf83295f5b6..31055ec1d1530ff5fe9c89e89957fdd6e83f7efc 100644
--- a/profiler/src/profile_conv_fwd_bias_relu.cpp
+++ b/profiler/src/profile_conv_fwd_bias_relu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_conv_fwd_bias_relu_add.cpp b/profiler/src/profile_conv_fwd_bias_relu_add.cpp
index 408dd02f78dd6241b5f0a67ae7aac4420c891943..8c2439a0c74be78b2d6ffcfdcce033c9ec0c7df5 100644
--- a/profiler/src/profile_conv_fwd_bias_relu_add.cpp
+++ b/profiler/src/profile_conv_fwd_bias_relu_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm.cpp b/profiler/src/profile_gemm.cpp
index 61bae6ae70ebcf965618a100c3f553de73182118..b3587ea98d8555f86344621f3c8d48967b98ef46 100644
--- a/profiler/src/profile_gemm.cpp
+++ b/profiler/src/profile_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_add_add_fastgelu.cpp b/profiler/src/profile_gemm_add_add_fastgelu.cpp
index c3c0fb7b67daf131ab53194bbf28a12aeedf62aa..8af3768a48aa9905700b55ab73e83777f20f8223 100644
--- a/profiler/src/profile_gemm_add_add_fastgelu.cpp
+++ b/profiler/src/profile_gemm_add_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_add_fastgelu.cpp b/profiler/src/profile_gemm_add_fastgelu.cpp
index 380b25a614c58d42209135218cca331de3bebe33..a09bb8340d3268ed642b1cd15f318f9458867c99 100644
--- a/profiler/src/profile_gemm_add_fastgelu.cpp
+++ b/profiler/src/profile_gemm_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_add_multiply.cpp b/profiler/src/profile_gemm_add_multiply.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..560467c264f2bcc8e406ed02de742c06942e27e0
--- /dev/null
+++ b/profiler/src/profile_gemm_add_multiply.cpp
@@ -0,0 +1,158 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_gemm_add_multiply_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+#define OP_NAME "gemm_add_multiply"
+#define OP_DESC "GEMM+Add+MULTIPLY"
+
+int profile_gemm_add_multiply(int argc, char* argv[])
+{
+    enum struct MatrixLayout
+    {
+        MK_KN_MN_MN_MN, // 0
+        MK_NK_MN_MN_MN, // 1
+        KM_KN_MN_MN_MN, // 2
+        KM_NK_MN_MN_MN, // 3
+    };
+
+    enum struct MatrixDataType
+    {
+        F32_F32_F32_F32_F32,      // 0
+        F16_F16_F16_F16_F16,      // 1
+        BF16_BF16_BF16_BF16_BF16, // 2
+        INT8_INT8_INT8_INT8_INT8, // 3
+    };
+
+    if(argc != 16)
+    {
+        // clang-format off
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n");
+        printf("arg3: matrix layout (0: E[m, n] = AddMultiply((A[m, k] * B[k, n] + D0[m, n]) x D1[m, n]);\n");
+        printf("                     1: E[m, n] = AddMultiply((A[m, k] * B[k, n] + D0[m, n]) x D1[m, n]);\n");
+        printf("                     2: E[m, n] = AddMultiply((A[m, k] * B[k, n] + D0[m, n]) x D1[m, n]);\n");
+        printf("                     3: E[m, n] = AddMultiply((A[m, k] * B[k, n] + D0[m, n]) x D1[m, n]))\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=no, 1=yes)\n");
+        printf("arg8 to 15: M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideE\n");
+        // clang-format on
+        exit(1);
+    }
+
+    const auto data_type       = static_cast<MatrixDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<MatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+
+    const int M = std::stoi(argv[8]);
+    const int N = std::stoi(argv[9]);
+    const int K = std::stoi(argv[10]);
+
+    const int StrideA  = std::stoi(argv[11]);
+    const int StrideB  = std::stoi(argv[12]);
+    const int StrideD0 = std::stoi(argv[13]);
+    const int StrideD1 = std::stoi(argv[14]);
+    const int StrideE  = std::stoi(argv[15]);
+
+    using F16 = ck::half_t;
+    using F32 = float;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    auto profile = [&](auto a_type,
+                       auto b_type,
+                       auto acc_type,
+                       auto d0_type,
+                       auto d1_type,
+                       auto e_type,
+                       auto a_layout,
+                       auto b_layout,
+                       auto d0_layout,
+                       auto d1_layout,
+                       auto e_layout) {
+        using ADataType   = decltype(a_type);
+        using BDataType   = decltype(b_type);
+        using AccDataType = decltype(acc_type);
+        using D0DataType  = decltype(d0_type);
+        using D1DataType  = decltype(d1_type);
+        using EDataType   = decltype(e_type);
+
+        using ALayout  = decltype(a_layout);
+        using BLayout  = decltype(b_layout);
+        using D0Layout = decltype(d0_layout);
+        using D1Layout = decltype(d1_layout);
+        using ELayout  = decltype(e_layout);
+
+        const int DefaultStrideA  = ck::is_same_v<ALayout, Row> ? K : M;
+        const int DefaultStrideB  = ck::is_same_v<BLayout, Row> ? N : K;
+        const int DefaultStrideD0 = ck::is_same_v<D0Layout, Row> ? N : M;
+        const int DefaultStrideD1 = ck::is_same_v<D1Layout, Row> ? N : M;
+        const int DefaultStrideE  = ck::is_same_v<ELayout, Row> ? N : M;
+
+        bool pass = ck::profiler::profile_gemm_add_multiply_impl<ADataType,
+                                                                 BDataType,
+                                                                 AccDataType,
+                                                                 D0DataType,
+                                                                 D1DataType,
+                                                                 EDataType,
+                                                                 ALayout,
+                                                                 BLayout,
+                                                                 D0Layout,
+                                                                 D1Layout,
+                                                                 ELayout>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? DefaultStrideA : StrideA,
+            (StrideB < 0) ? DefaultStrideB : StrideB,
+            (StrideD0 < 0) ? DefaultStrideD0 : StrideD0,
+            (StrideD1 < 0) ? DefaultStrideD1 : StrideD1,
+            (StrideE < 0) ? DefaultStrideE : StrideE);
+
+        return pass ? 0 : 1;
+    };
+
+    if(data_type == MatrixDataType::F16_F16_F16_F16_F16 && layout == MatrixLayout::MK_KN_MN_MN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, F16{}, F16{}, Row{}, Row{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::F16_F16_F16_F16_F16 &&
+            layout == MatrixLayout::MK_NK_MN_MN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, F16{}, F16{}, Row{}, Col{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::F16_F16_F16_F16_F16 &&
+            layout == MatrixLayout::KM_KN_MN_MN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, F16{}, F16{}, Col{}, Row{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::F16_F16_F16_F16_F16 &&
+            layout == MatrixLayout::KM_NK_MN_MN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, F16{}, F16{}, Col{}, Col{}, Row{}, Row{}, Row{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_gemm_add_multiply);
diff --git a/profiler/src/profile_gemm_add_relu_add_layernorm.cpp b/profiler/src/profile_gemm_add_relu_add_layernorm.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..558d255ce110565e90f6eff5fae86ce6bf6447bb
--- /dev/null
+++ b/profiler/src/profile_gemm_add_relu_add_layernorm.cpp
@@ -0,0 +1,215 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_gemm_add_relu_add_layernorm_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+#define OP_NAME "gemm_add_relu_add_layernorm"
+#define OP_DESC "GEMM+Add+Relu+Add+Layernorm"
+
+int profile_gemm_add_relu_add_layernorm(int argc, char* argv[])
+{
+    enum struct MatrixLayout
+    {
+        MK_KN_MN_MN_MN, // 0
+        MK_NK_MN_MN_MN, // 1
+        KM_KN_MN_MN_MN, // 2
+        KM_NK_MN_MN_MN, // 3
+    };
+
+    enum struct MatrixDataType
+    {
+        F32,  // 0
+        F16,  // 1
+        BF16, // 2
+    };
+
+    if(argc != 16)
+    {
+        // clang-format off
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16)\n");
+        printf("arg3: matrix layout (0: H[m, n] = Layernorm(Relu(A[m, k] * B[k, n] + D0[m, n]) + D1[m, n]);\n");
+        printf("                     1: H[m, n] = Layernorm(Relu(A[m, k] * B[n, k] + D0[m, n]) + D1[m, n]);\n");
+        printf("                     2: H[m, n] = Layernorm(Relu(A[k, m] * B[k, n] + D0[m, n]) + D1[m, n]);\n");
+        printf("                     3: H[m, n] = Layernorm(Relu(A[k, m] * B[n, k] + D0[m, n]) + D1[m, n]))\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=no, 1=yes)\n");
+        printf("arg8 to 15: M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideH\n");
+        // clang-format on
+        exit(1);
+    }
+
+    const auto data_type       = static_cast<MatrixDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<MatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+
+    const int M = std::stoi(argv[8]);
+    const int N = std::stoi(argv[9]);
+    const int K = std::stoi(argv[10]);
+
+    const int StrideA  = std::stoi(argv[11]);
+    const int StrideB  = std::stoi(argv[12]);
+    const int StrideD0 = std::stoi(argv[13]);
+    const int StrideD1 = std::stoi(argv[14]);
+    const int StrideH  = std::stoi(argv[15]);
+
+    using F16 = ck::half_t;
+    using F32 = float;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    auto profile = [&](auto a_type,
+                       auto b_type,
+                       auto acc_type,
+                       auto d0_type,
+                       auto d1_type,
+                       auto e_mean_var_type,
+                       auto gamma_type,
+                       auto beta_type,
+                       auto h_type,
+                       auto a_layout,
+                       auto b_layout,
+                       auto d0_layout,
+                       auto d1_layout,
+                       auto h_layout) {
+        using ADataType        = decltype(a_type);
+        using BDataType        = decltype(b_type);
+        using AccDataType      = decltype(acc_type);
+        using D0DataType       = decltype(d0_type);
+        using D1DataType       = decltype(d1_type);
+        using EMeanVarDataType = decltype(e_mean_var_type);
+        using GammaDataType    = decltype(gamma_type);
+        using BetaDataType     = decltype(beta_type);
+        using HDataType        = decltype(h_type);
+
+        using ALayout  = decltype(a_layout);
+        using BLayout  = decltype(b_layout);
+        using D0Layout = decltype(d0_layout);
+        using D1Layout = decltype(d1_layout);
+        using HLayout  = decltype(h_layout);
+
+        const int DefaultStrideA  = ck::is_same_v<ALayout, Row> ? K : M;
+        const int DefaultStrideB  = ck::is_same_v<BLayout, Row> ? N : K;
+        const int DefaultStrideD0 = ck::is_same_v<D0Layout, Row> ? N : M;
+        const int DefaultStrideD1 = ck::is_same_v<D1Layout, Row> ? N : M;
+        const int DefaultStrideH  = ck::is_same_v<HLayout, Row> ? N : M;
+
+        bool pass = ck::profiler::profile_gemm_add_relu_add_layernorm_impl<ADataType,
+                                                                           BDataType,
+                                                                           AccDataType,
+                                                                           D0DataType,
+                                                                           D1DataType,
+                                                                           EMeanVarDataType,
+                                                                           GammaDataType,
+                                                                           BetaDataType,
+                                                                           HDataType,
+                                                                           ALayout,
+                                                                           BLayout,
+                                                                           D0Layout,
+                                                                           D1Layout,
+                                                                           HLayout>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? DefaultStrideA : StrideA,
+            (StrideB < 0) ? DefaultStrideB : StrideB,
+            (StrideD0 < 0) ? DefaultStrideD0 : StrideD0,
+            (StrideD1 < 0) ? DefaultStrideD1 : StrideD1,
+            (StrideH < 0) ? DefaultStrideH : StrideH);
+
+        return pass ? 0 : 1;
+    };
+
+    if(data_type == MatrixDataType::F16 && layout == MatrixLayout::MK_KN_MN_MN_MN)
+    {
+        return profile(F16{},
+                       F16{},
+                       F32{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       Row{},
+                       Row{},
+                       Row{},
+                       Row{},
+                       Row{});
+    }
+    else if(data_type == MatrixDataType::F16 && layout == MatrixLayout::MK_NK_MN_MN_MN)
+    {
+        return profile(F16{},
+                       F16{},
+                       F32{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       Row{},
+                       Col{},
+                       Row{},
+                       Row{},
+                       Row{});
+    }
+    else if(data_type == MatrixDataType::F16 && layout == MatrixLayout::KM_KN_MN_MN_MN)
+    {
+        return profile(F16{},
+                       F16{},
+                       F32{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       Col{},
+                       Row{},
+                       Row{},
+                       Row{},
+                       Row{});
+    }
+    else if(data_type == MatrixDataType::F16 && layout == MatrixLayout::KM_NK_MN_MN_MN)
+    {
+        return profile(F16{},
+                       F16{},
+                       F32{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       F16{},
+                       Col{},
+                       Col{},
+                       Row{},
+                       Row{},
+                       Row{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_gemm_add_relu_add_layernorm);
diff --git a/profiler/src/profile_gemm_bias_add_reduce.cpp b/profiler/src/profile_gemm_bias_add_reduce.cpp
index 6d86db08223a771ddc4373c014a6a2a42f92285b..76daffbc67cb95ac19d52a6791d90816a5e6656b 100644
--- a/profiler/src/profile_gemm_bias_add_reduce.cpp
+++ b/profiler/src/profile_gemm_bias_add_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_bilinear.cpp b/profiler/src/profile_gemm_bilinear.cpp
index 3480014ba6e6a2088c1ee1a221eb99eee6191241..a1a48616b42cac2f0babf59955a68cc49c323ac3 100644
--- a/profiler/src/profile_gemm_bilinear.cpp
+++ b/profiler/src/profile_gemm_bilinear.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_fastgelu.cpp b/profiler/src/profile_gemm_fastgelu.cpp
index 2a137224cb096f4bf2968f06be13d71d44614794..93573002ef1c367a00d42dc2ea0fe99fca98086a 100644
--- a/profiler/src/profile_gemm_fastgelu.cpp
+++ b/profiler/src/profile_gemm_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_reduce.cpp b/profiler/src/profile_gemm_reduce.cpp
index 395bf0627e43617d04d30a903a1f3cbdecf4a557..48f6f5eb49da151136600c2d03b99d959d51565f 100644
--- a/profiler/src/profile_gemm_reduce.cpp
+++ b/profiler/src/profile_gemm_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_splitk.cpp b/profiler/src/profile_gemm_splitk.cpp
index f636ce718c669feb23a60fb1853a1219fea3da55..cc2da73cb72704ffb2c1283976f6307b2e89908c 100644
--- a/profiler/src/profile_gemm_splitk.cpp
+++ b/profiler/src/profile_gemm_splitk.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_grouped_conv_bwd_data.cpp b/profiler/src/profile_grouped_conv_bwd_data.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..351bb72c8153ffaea6359b38d344fec5efe043bc
--- /dev/null
+++ b/profiler/src/profile_grouped_conv_bwd_data.cpp
@@ -0,0 +1,157 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+namespace {
+
+enum struct ConvLayout
+{
+    GNHWC_GKYXC_GNHWK, // 0
+    NHWGC_GKYXC_NHWGK, // 1
+};
+
+enum struct ConvDataType
+{
+    F32_F32_F32,    // 0
+    F16_F16_F16,    // 1
+    BF16_BF16_BF16, // 2
+};
+
+#define OP_NAME "grouped_conv_bwd_data"
+#define OP_DESC "Grouped Convolution Backward Data"
+
+static void print_helper_msg()
+{
+    std::cout
+        // clang-format off
+        << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+        << "arg2: data type (0: Output fp32, Weight fp32, Input fp32\n"
+        << "                 1: Output fp16, Weight fp16, Input fp16\n"
+        << "                 2: Output bf16, Weight bf16, Input bf16\n"
+        << "arg3: tensor layout (0: Output[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Input[G, N, Ho, Wo, K]\n"
+        << "                     1: Output[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Input[N, Ho, Wo, G, K])\n"
+        << "arg4: verification (0: no, 1: yes)\n"
+        << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
+        << "arg6: print tensor value (0: no; 1: yes)\n"
+        << "arg7: time kernel (0: no, 1: yes)\n"
+        << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+    // clang-format on
+}
+
+} // namespace
+
+int profile_grouped_conv_bwd_data(int argc, char* argv[])
+{
+    // 8 for control, 1 for num_dim_spatial
+    if(argc < 9)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto data_type       = static_cast<ConvDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<ConvLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+    const int num_dim_spatial  = std::stoi(argv[8]);
+
+    // 8 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
+    if(argc != 8 + 1 + 4 + 6 * num_dim_spatial)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto params = ck::utils::conv::parse_conv_param(num_dim_spatial, 9, argv);
+
+    using F32  = float;
+    using F16  = ck::half_t;
+    using BF16 = ck::bhalf_t;
+
+    using GNHWC = ck::tensor_layout::convolution::GNHWC;
+    using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+    using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+    using GNHWK = ck::tensor_layout::convolution::GNHWK;
+    using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+    constexpr auto I2 = ck::Number<2>{};
+
+    auto profile = [&](auto num_dim_spatial_tmp,
+                       auto out_layout,
+                       auto wei_layout,
+                       auto in_layout,
+                       auto wei_type,
+                       auto out_type,
+                       auto in_type) {
+        constexpr ck::index_t NDimSpatial = num_dim_spatial_tmp.value;
+
+        using OutLayout = decltype(out_layout);
+        using WeiLayout = decltype(wei_layout);
+        using InLayout  = decltype(in_layout);
+
+        using OutDataType = decltype(out_type);
+        using WeiDataType = decltype(wei_type);
+        using InDataType  = decltype(in_type);
+
+        bool pass = ck::profiler::profile_grouped_conv_bwd_data_impl<NDimSpatial,
+                                                                     OutLayout,
+                                                                     WeiLayout,
+                                                                     InLayout,
+                                                                     OutDataType,
+                                                                     WeiDataType,
+                                                                     InDataType>(
+            do_verification, init_method, do_log, time_kernel, params);
+
+        return pass ? 0 : 1;
+    };
+
+    // GNHWC_GKYXC_GNHWK
+    if(num_dim_spatial == 2 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_BF16_BF16)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, BF16{}, BF16{}, BF16{});
+        }
+    }
+    // NHWGC_GKYXC_NHWGK
+    else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_BF16_BF16)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, BF16{}, BF16{}, BF16{});
+        }
+    }
+
+    std::cout << "this data_type & layout is not implemented" << std::endl;
+
+    return 1;
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_grouped_conv_bwd_data);
diff --git a/profiler/src/profile_grouped_conv_bwd_weight.cpp b/profiler/src/profile_grouped_conv_bwd_weight.cpp
index dfd8a099f5408f76fd51a9c69a988591edf5348f..d8edcf0f657988b7fea0191efa67af507fe06414 100644
--- a/profiler/src/profile_grouped_conv_bwd_weight.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_weight.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <initializer_list>
@@ -15,6 +15,7 @@ enum struct ConvLayout
 {
     GNCHW_GKCYX_GNKHW, // 0
     GNHWC_GKYXC_GNHWK, // 1
+    NHWGC_GKYXC_NHWGK, // 2
 };
 
 enum struct ConvDataType
@@ -37,6 +38,8 @@ static void print_helper_msg()
                  "N, K, Ho, Wo]\n"
               << "                     1: Input[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Output[G, "
                  "N, Ho, Wo, K]\n"
+              << "                     2: Input[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Output[N, "
+                 "Ho, Wo, G, K]\n"
               << "arg4: verification (0: no, 1: yes)\n"
               << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
               << "arg6: print tensor value (0: no; 1: yes)\n"
@@ -82,6 +85,7 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
 
     using GNWC   = ck::tensor_layout::convolution::GNWC;
     using GNHWC  = ck::tensor_layout::convolution::GNHWC;
+    using NHWGC  = ck::tensor_layout::convolution::NHWGC;
     using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
 
     using GKXC   = ck::tensor_layout::convolution::GKXC;
@@ -90,6 +94,7 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
 
     using GNWK   = ck::tensor_layout::convolution::GNWK;
     using GNHWK  = ck::tensor_layout::convolution::GNHWK;
+    using NHWGK  = ck::tensor_layout::convolution::NHWGK;
     using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
 
     constexpr auto I1 = ck::Number<1>{};
@@ -157,6 +162,22 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
             return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, BF16{}, F32{}, BF16{});
         }
     }
+    else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_F32_BF16)
+        {
+            // fp32 atomic add is used for weight tensor in bf16 kernel
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, BF16{}, F32{}, BF16{});
+        }
+    }
     else if(num_dim_spatial == 3 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
diff --git a/profiler/src/profile_grouped_conv_fwd.cpp b/profiler/src/profile_grouped_conv_fwd.cpp
index 9ff3c15af05a5f67988ef4842e00e2671289bb1b..d0b424cde64249aff7da0646a6bb11e83ccecf39 100644
--- a/profiler/src/profile_grouped_conv_fwd.cpp
+++ b/profiler/src/profile_grouped_conv_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_grouped_gemm.cpp b/profiler/src/profile_grouped_gemm.cpp
index 65e24bd9cc1fba07ab17010c1a95fdd02d991dfe..d023db54de47cdb4464148985547a80ad441190b 100644
--- a/profiler/src/profile_grouped_gemm.cpp
+++ b/profiler/src/profile_grouped_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -29,6 +29,11 @@ enum struct GemmDataType
     INT8_INT8_INT8, // 3
 };
 
+#define OP_NAME "grouped_gemm"
+#define OP_DESC "Grouped GEMM"
+
+namespace {
+
 std::vector<int> argToIntArray(char* input)
 {
     std::vector<int> out;
@@ -45,25 +50,26 @@ std::vector<int> argToIntArray(char* input)
     return out;
 }
 
-#define OP_NAME "grouped_gemm"
-#define OP_DESC "Grouped GEMM"
-
 int profile_grouped_gemm(int argc, char* argv[])
 {
-    if(!(argc == 14))
+    if(argc < 14)
     {
-        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
-        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n");
-        printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
-        printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
-        printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
-        printf("                     3: A[k, m] * B[n, k] = C[m, n])\n");
-        printf("arg4: verification (0: no; 1: yes)\n");
-        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
-        printf("arg6: print tensor value (0: no; 1: yes)\n");
-        printf("arg7: time kernel (0=n0, 1=yes)\n");
-        printf("arg8 to 13: Ms, Ns, Ks, StrideAs, StrideBs, StrideCs (e.g., 256,256 128,128 64,64 "
-               "64,64 64,64 128,128)\n");
+        std::cout
+            << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+            << "arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n"
+            << "arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n"
+            << "                     1: A[m, k] * B[n, k] = C[m, n];\n"
+            << "                     2: A[k, m] * B[k, n] = C[m, n];\n"
+            << "                     3: A[k, m] * B[n, k] = C[m, n])\n"
+            << "arg4: verification (0: no; 1: yes)\n"
+            << "arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+            << "arg6: print tensor value (0: no; 1: yes)\n"
+            << "arg7: time kernel (0=n0, 1=yes)\n"
+            << "arg8 to 13: Ms, Ns, Ks, StrideAs, StrideBs, StrideCs (e.g., 256,256 128,128 64,64 "
+               "64,64 64,64 128,128)\n"
+            << "arg15: kbatch value (default 4)\n"
+            << std::endl;
+
         exit(1);
     }
 
@@ -81,6 +87,7 @@ int profile_grouped_gemm(int argc, char* argv[])
     const auto StrideAs = argToIntArray(argv[11]);
     const auto StrideBs = argToIntArray(argv[12]);
     const auto StrideCs = argToIntArray(argv[13]);
+    const int kbatch    = argc == 15 ? std::stoi(argv[14]) : 1;
 
     if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
@@ -99,7 +106,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
     {
@@ -118,7 +126,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
     {
@@ -137,7 +146,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
     {
@@ -156,7 +166,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else
     {
@@ -166,4 +177,6 @@ int profile_grouped_gemm(int argc, char* argv[])
     return 0;
 }
 
+} // anonymous namespace
+
 REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_grouped_gemm);
diff --git a/profiler/src/profile_grouped_gemm_fastgelu.cpp b/profiler/src/profile_grouped_gemm_fastgelu.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..50ecf25caebdedb385d9ab523abe5f73f1afb8d5
--- /dev/null
+++ b/profiler/src/profile_grouped_gemm_fastgelu.cpp
@@ -0,0 +1,177 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_grouped_gemm_fastgelu_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+enum struct GemmMatrixLayout
+{
+    MK_KN_MN, // 0
+    MK_NK_MN, // 1
+    KM_KN_MN, // 2
+    KM_NK_MN, // 3
+    MK_KN_NM, // 4
+    MK_NK_NM, // 5
+    KM_KN_NM, // 6
+    KM_NK_NM, // 7
+};
+
+enum struct GemmDataType
+{
+    F32_F32_F32,    // 0
+    F16_F16_F16,    // 1
+    BF16_BF16_BF16, // 2
+    INT8_INT8_INT8, // 3
+};
+
+#define OP_NAME "grouped_gemm_fastgelu"
+#define OP_DESC "Grouped GEMM+FastGelu"
+
+namespace {
+
+std::vector<int> argToIntArray(char* input)
+{
+    std::vector<int> out;
+
+    std::istringstream in(input);
+
+    std::string item;
+
+    while(std::getline(in, item, ','))
+    {
+        out.push_back(std::stoi(item));
+    }
+
+    return out;
+}
+
+int profile_grouped_gemm_fastgelu(int argc, char* argv[])
+{
+    if(!(argc == 14))
+    {
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n");
+        printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
+        printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
+        printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
+        printf("                     3: A[k, m] * B[n, k] = C[m, n])\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=n0, 1=yes)\n");
+        printf("arg8 to 13: Ms, Ns, Ks, StrideAs, StrideBs, StrideCs (e.g., 256,256 128,128 64,64 "
+               "64,64 64,64 128,128)\n");
+        exit(1);
+    }
+
+    const auto data_type       = static_cast<GemmDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<GemmMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+
+    const auto Ms = argToIntArray(argv[8]);
+    const auto Ns = argToIntArray(argv[9]);
+    const auto Ks = argToIntArray(argv[10]);
+
+    const auto StrideAs = argToIntArray(argv[11]);
+    const auto StrideBs = argToIntArray(argv[12]);
+    const auto StrideCs = argToIntArray(argv[13]);
+
+    if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        ck::profiler::profile_grouped_gemm_fastgelu_impl<ck::half_t,
+                                                         ck::half_t,
+                                                         ck::half_t,
+                                                         float,
+                                                         ck::tensor_layout::gemm::RowMajor,
+                                                         ck::tensor_layout::gemm::RowMajor,
+                                                         ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            Ms,
+            Ns,
+            Ks,
+            StrideAs,
+            StrideBs,
+            StrideCs);
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        ck::profiler::profile_grouped_gemm_fastgelu_impl<ck::half_t,
+                                                         ck::half_t,
+                                                         ck::half_t,
+                                                         float,
+                                                         ck::tensor_layout::gemm::RowMajor,
+                                                         ck::tensor_layout::gemm::ColumnMajor,
+                                                         ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            Ms,
+            Ns,
+            Ks,
+            StrideAs,
+            StrideBs,
+            StrideCs);
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        ck::profiler::profile_grouped_gemm_fastgelu_impl<ck::half_t,
+                                                         ck::half_t,
+                                                         ck::half_t,
+                                                         float,
+                                                         ck::tensor_layout::gemm::ColumnMajor,
+                                                         ck::tensor_layout::gemm::RowMajor,
+                                                         ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            Ms,
+            Ns,
+            Ks,
+            StrideAs,
+            StrideBs,
+            StrideCs);
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        ck::profiler::profile_grouped_gemm_fastgelu_impl<ck::half_t,
+                                                         ck::half_t,
+                                                         ck::half_t,
+                                                         float,
+                                                         ck::tensor_layout::gemm::ColumnMajor,
+                                                         ck::tensor_layout::gemm::ColumnMajor,
+                                                         ck::tensor_layout::gemm::RowMajor>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            Ms,
+            Ns,
+            Ks,
+            StrideAs,
+            StrideBs,
+            StrideCs);
+    }
+    else
+    {
+        throw std::runtime_error("wrong! this GEMM data_type & layout is not implemented");
+    }
+
+    return 0;
+}
+
+} // anonymous namespace
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_grouped_gemm_fastgelu);
diff --git a/profiler/src/profile_groupnorm.cpp b/profiler/src/profile_groupnorm.cpp
index 2741f52717a5b7e370be2ddf3420273fc78df0e5..d55784ff0ad7274fce9bbf66b26a508f5994a578 100644
--- a/profiler/src/profile_groupnorm.cpp
+++ b/profiler/src/profile_groupnorm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
@@ -64,7 +64,7 @@ int profile_groupnorm(int argc, char* argv[])
     ck::DataTypeEnum data_type  = ck::DataTypeEnum::Half;
     bool do_verification        = false;
     int init_method             = 0;
-    bool do_log                 = 0;
+    bool do_log                 = false;
     bool time_kernel            = 1;
     std::vector<index_t> length = {64, 16, 16, 32, 40};
 
diff --git a/profiler/src/profile_layernorm.cpp b/profiler/src/profile_layernorm.cpp
index e93fc2dbd2bf004389eed968e0518a2cdb1ea4f6..7bf210e67837fa8fe0fe8982c40a1793344823a0 100644
--- a/profiler/src/profile_layernorm.cpp
+++ b/profiler/src/profile_layernorm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_max_pool3d_fwd.cpp b/profiler/src/profile_max_pool3d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..cf6db2cfc91e1b7d52af6cc7e9f9d43f4051599e
--- /dev/null
+++ b/profiler/src/profile_max_pool3d_fwd.cpp
@@ -0,0 +1,168 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+#include <unordered_map>
+
+#include "profiler/data_type_enum.hpp"
+#include "profiler/profile_pool3d_fwd_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+using ck::index_t;
+
+struct maxPoolFwdArgParser
+{
+    std::unordered_map<std::string, std::vector<int>> long_opts = {
+        {"length", {}}, {"wsize", {}}, {"wstride", {}}, {"pad1", {}}, {"pad2", {}}};
+
+    bool parse_opt(int argc, char* argv[], const std::string& key, int i)
+    {
+        if(std::string("--") + key == argv[i])
+        {
+            int pos = i;
+            while(++i < argc && argv[i][0] != '-') {}
+            int end = i;
+            for(int j = pos + 1; j < end; j++)
+            {
+                long_opts[key].push_back(std::stoi(argv[j]));
+            }
+            return true;
+        }
+        return false;
+    }
+
+    void operator()(int argc, char* argv[])
+    {
+        for(auto& kv : long_opts)
+        {
+            for(int i = 1; i < argc; i++)
+            {
+                if(parse_opt(argc, argv, kv.first, i))
+                    break;
+            }
+        }
+    }
+};
+
+void print_help_max_pool3d_fwd()
+{
+    std::cout << "arg1: data type (0: fp16; 1: fp32)\n"
+              << "arg2: verification (0: no; 1: yes)\n"
+              << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+              << "arg4: print tensor value (0: no; 1: yes)\n"
+              << "arg5: time kernel (0=no, 1=yes)\n"
+              << "arg6: return index (0=no, 1=yes)\n"
+              << "--length: input tensor length for NCDHW(e.g, --length 2 32 30 30 30) \n"
+              << "--wsize: window size for ZYX (e.g, --wsize 2 2 2) \n"
+              << "--wstride: window stride for DHW (e.g, --wstride 2 2 2) \n"
+              << "--pad1: left side of padding in DHW (e.g, --pad1 1 1 1) \n"
+              << "--pad2: right side of padding in DHW (e.g, --pad2 1 1 1) \n"
+              << "eg: ckProfiler max_pool3d_fwd 0 1 2 0 1 0 --length 2 32 30 30 30 --wsize 2 2 2 "
+                 "--wstride 2 2 2 --pad1 1 1 1 --pad2 1 1 1"
+              << std::endl;
+}
+
+int profile_max_pool3d_fwd(int argc, char* argv[])
+{
+    ck::DataTypeEnum data_type = ck::DataTypeEnum::Half;
+    bool do_verification       = true;
+    int init_method            = 0;
+    bool do_log                = false;
+    bool time_kernel           = true;
+    bool return_index          = false;
+
+    std::vector<index_t> in_length = {2, 32, 30, 30, 30};
+    std::vector<index_t> wsize     = {2, 2, 2};
+    std::vector<index_t> wstride   = {2, 2, 2};
+    std::vector<index_t> pad1      = {1, 1, 1};
+    std::vector<index_t> pad2      = {1, 1, 1};
+
+    if(argc != 2 && argc != 30)
+    {
+        print_help_max_pool3d_fwd();
+        return 0;
+    }
+    else if(argc == 30)
+    {
+        data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
+        do_verification = std::stoi(argv[3]);
+        init_method     = std::stoi(argv[4]);
+        do_log          = std::stoi(argv[5]);
+        time_kernel     = std::stoi(argv[6]);
+        return_index    = std::stoi(argv[7]);
+
+        // parse the long options
+        maxPoolFwdArgParser arg_parser;
+        arg_parser(argc, argv);
+        in_length = arg_parser.long_opts["length"];
+        wsize     = arg_parser.long_opts["wsize"];
+        wstride   = arg_parser.long_opts["wstride"];
+        pad1      = arg_parser.long_opts["pad1"];
+        pad2      = arg_parser.long_opts["pad2"];
+    }
+
+    using F16                 = ck::half_t;
+    using F32                 = float;
+    using I32                 = int32_t;
+    constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+    if(data_type == ck::DataTypeEnum::Half)
+    {
+        if(return_index)
+            ck::profiler::profile_pool3d_fwd_impl<F16, F16, F16, I32, ReduceOpId, false, true>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+        else
+            ck::profiler::profile_pool3d_fwd_impl<F16, F16, F16, I32, ReduceOpId, false, false>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+    }
+    else if(data_type == ck::DataTypeEnum::Float)
+    {
+        if(return_index)
+            ck::profiler::profile_pool3d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, true>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+        else
+            ck::profiler::profile_pool3d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, false>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+    }
+    else
+    {
+        throw std::runtime_error("not implemented yet");
+    }
+
+    return 0;
+}
+
+REGISTER_PROFILER_OPERATION("max_pool3d_fwd", "max_pool3d fwd", profile_max_pool3d_fwd);
diff --git a/profiler/src/profile_reduce.cpp b/profiler/src/profile_reduce.cpp
index 6925371858ee913dcb5d2b529c3e350a6eec5f4f..e4af5680a5d8a192edfc380d18fbe0a2c595cbbb 100644
--- a/profiler/src/profile_reduce.cpp
+++ b/profiler/src/profile_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <fstream>
diff --git a/profiler/src/profile_softmax.cpp b/profiler/src/profile_softmax.cpp
index 30f627dd29e57f3fda5bef57353f943fca621701..dfe8d95c904e297fcd2f7b83a166fc61c8950233 100644
--- a/profiler/src/profile_softmax.cpp
+++ b/profiler/src/profile_softmax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
@@ -92,27 +92,76 @@ int profile_softmax(int argc, char* argv[])
     {
         if(data_type == SoftmaxDataType::F16_F16)
         {
-            ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3>(do_verification,
-                                                                                 init_method,
-                                                                                 do_log,
-                                                                                 time_kernel,
-                                                                                 length,
-                                                                                 stride,
-                                                                                 reduce,
-                                                                                 float(alpha),
-                                                                                 float(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3, 1>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3, 2>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3, 3>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else if(data_type == SoftmaxDataType::F32_F32)
         {
-            ck::profiler::profile_softmax_impl<float, float, float, 3>(do_verification,
-                                                                       init_method,
-                                                                       do_log,
-                                                                       time_kernel,
-                                                                       length,
-                                                                       stride,
-                                                                       reduce,
-                                                                       float(alpha),
-                                                                       float(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<float, float, float, 3, 1>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<float, float, float, 3, 2>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<float, float, float, 3, 3>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else
         {
@@ -124,27 +173,97 @@ int profile_softmax(int argc, char* argv[])
     {
         if(data_type == SoftmaxDataType::F16_F16)
         {
-            ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4>(do_verification,
-                                                                                 init_method,
-                                                                                 do_log,
-                                                                                 time_kernel,
-                                                                                 length,
-                                                                                 stride,
-                                                                                 reduce,
-                                                                                 float(alpha),
-                                                                                 float(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 1>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 2>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 3>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 4)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 4>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else if(data_type == SoftmaxDataType::F32_F32)
         {
-            ck::profiler::profile_softmax_impl<float, float, float, 4>(do_verification,
-                                                                       init_method,
-                                                                       do_log,
-                                                                       time_kernel,
-                                                                       length,
-                                                                       stride,
-                                                                       reduce,
-                                                                       float(alpha),
-                                                                       float(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 1>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 2>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 3>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 4)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 4>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else
         {
diff --git a/profiler/src/profiler.cpp b/profiler/src/profiler.cpp
index 080117e390c4df1bc14195649382267bcccc18f7..0f528c008f1d17afd42139d572d4f1cf1963fadd 100644
--- a/profiler/src/profiler.cpp
+++ b/profiler/src/profiler.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/profiler/src/profiler_operation_registry.hpp b/profiler/src/profiler_operation_registry.hpp
index 91ff291233066527609133dc155e67388119385a..276b7b38dcd28ee935c108819cd0102b2bec4862 100644
--- a/profiler/src/profiler_operation_registry.hpp
+++ b/profiler/src/profiler_operation_registry.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <iostream>
diff --git a/script/check_copyright_year.sh b/script/check_copyright_year.sh
new file mode 100755
index 0000000000000000000000000000000000000000..f7709472efaa2f8bba4a5d7807b98a66f8caa807
--- /dev/null
+++ b/script/check_copyright_year.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+
+current_year=$(date +%Y)
+exit_code=0
+
+for file in $@; do
+    if grep -q "Copyright (c)" $file
+    then
+        if ! grep -q "Copyright (c).*$current_year" $file
+        then
+            echo "ERROR: File $file has a copyright notice without the current year ($current_year)."
+            exit_code=1
+        fi
+    fi
+done
+
+exit $exit_code
diff --git a/script/cmake-ck-dev.sh b/script/cmake-ck-dev.sh
index 2e605ce8def00add5c2390441ab7a7071d39f073..426f68d4432e19f378c543e1edb90364f97e54af 100755
--- a/script/cmake-ck-dev.sh
+++ b/script/cmake-ck-dev.sh
@@ -8,12 +8,12 @@ MY_PROJECT_SOURCE=$1
 cmake                                                                                             \
 -D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
 -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
--D CMAKE_CXX_FLAGS="-O3 -ftemplate-backtrace-limit=0 -gline-tables-only -save-temps=$PWD"         \
+-D CMAKE_CXX_FLAGS="-std=c++17 -O3 -ftemplate-backtrace-limit=0  -fPIE  -Wno-gnu-line-marker      \
+-save-temps=$PWD"                                                                                 \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=ON                                                                                   \
--D GPU_TARGETS="gfx908;gfx90a"                                                                    \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
 
-#-D AMDGPU_TARGETS=gfx90a;gfx908
diff --git a/script/cmake-ck-release.sh b/script/cmake-ck-release.sh
index 268b1ebf9b7fa1743ef06b8cf3073dc44c0a7bc1..787eabbf96bc1029d54ff592bee8eaca6e214b89 100755
--- a/script/cmake-ck-release.sh
+++ b/script/cmake-ck-release.sh
@@ -11,9 +11,8 @@ cmake
 -D CMAKE_CXX_FLAGS="-O3"                                                                          \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=OFF                                                                                  \
--D GPU_TARGETS="gfx908;gfx90a"                                                                      \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
 
-#-D AMDGPU_TARGETS=gfx90a;gfx908
diff --git a/script/install_precommit.sh b/script/install_precommit.sh
new file mode 100755
index 0000000000000000000000000000000000000000..daa5cb29ae47f81c1e67d97f53caef2d54df6862
--- /dev/null
+++ b/script/install_precommit.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+
+run_and_check() {
+    "$@"
+    status=$?
+    if [ $status -ne 0 ]; then
+        echo "Error with \"$@\": Exited with status $status"
+        exit $status
+    fi
+    return $status
+}
+
+echo "I: Installing tools required for pre-commit checks..."
+run_and_check apt install clang-format-10
+
+echo "I: Installing pre-commit itself..."
+run_and_check pip3 install pre-commit
+run_and_check pre-commit install
+
+echo "I: Installation successful."
diff --git a/script/process_perf_data.py b/script/process_perf_data.py
index 638e4ef5644c7e814739893cfedabc6afeed28bb..e8b8e1458cc840978d22ea2d93347f7d06101aee 100644
--- a/script/process_perf_data.py
+++ b/script/process_perf_data.py
@@ -3,6 +3,7 @@ import os, io, argparse, datetime
 #import numpy as np
 import sqlalchemy
 from sqlalchemy.types import NVARCHAR, Float, Integer
+from sqlalchemy import text
 import pymysql
 import pandas as pd
 from sshtunnel import SSHTunnelForwarder
@@ -141,8 +142,8 @@ def parse_logfile(logfile):
 
 
 def get_baseline(table, connection):
-    query = '''SELECT * from '''+table+''' WHERE Datetime = (SELECT MAX(Datetime) FROM '''+table+''' where Branch_ID='develop' );'''
-    return pd.read_sql_query(query, connection)
+    query = text('''SELECT * from '''+table+''' WHERE Datetime = (SELECT MAX(Datetime) FROM '''+table+''' where Branch_ID='develop' );''')
+    return pd.read_sql(query, connection)
 
 def store_new_test_result(table_name, test_results, testlist, branch_name, node_id, gpu_arch, compute_units, rocm_vers, hip_vers, environment, connection):
     params=[str(branch_name),str(node_id),str(gpu_arch),compute_units,str(rocm_vers),str(hip_vers),str(environment),str(datetime.datetime.now())]
diff --git a/script/uninstall_precommit.sh b/script/uninstall_precommit.sh
new file mode 100755
index 0000000000000000000000000000000000000000..b0d4d15166fd85c711b0fb4f1bb87eb9ece720aa
--- /dev/null
+++ b/script/uninstall_precommit.sh
@@ -0,0 +1 @@
+pre-commit uninstall
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index bea05103098f7f5655f77e16b0696585b9b1ee7e..22589d975b6a568d6494ca8529def704983e6da1 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -27,7 +27,7 @@ function(add_gtest_executable TEST_NAME)
     # suppress gtest warnings
     target_compile_options(${TEST_NAME} PRIVATE -Wno-global-constructors -Wno-undef)
     target_link_libraries(${TEST_NAME} PRIVATE gtest_main)
-    add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}> )
+    add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}>)
     rocm_install(TARGETS ${TEST_NAME} COMPONENT tests)
 endfunction(add_gtest_executable TEST_NAME)
 
@@ -36,6 +36,7 @@ add_subdirectory(space_filling_curve)
 add_subdirectory(conv_util)
 add_subdirectory(reference_conv_fwd)
 add_subdirectory(gemm)
+add_subdirectory(gemm_layernorm)
 add_subdirectory(gemm_split_k)
 add_subdirectory(gemm_reduce)
 add_subdirectory(batched_gemm)
@@ -55,6 +56,10 @@ add_subdirectory(normalization)
 add_subdirectory(data_type)
 add_subdirectory(elementwise_normalization)
 add_subdirectory(batchnorm)
+add_subdirectory(contraction)
+add_subdirectory(pool_fwd)
+add_subdirectory(batched_gemm_multi_d)
+add_subdirectory(grouped_convnd_bwd_data)
 if(GPU_TARGETS MATCHES "gfx1100")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/batched_gemm/CMakeLists.txt b/test/batched_gemm/CMakeLists.txt
index 0574f98e872dcb436b14df3f536f06a787d28c8d..12142c33e71cc3751d4ac79cf739d02000da8731 100644
--- a/test/batched_gemm/CMakeLists.txt
+++ b/test/batched_gemm/CMakeLists.txt
@@ -1,15 +1,22 @@
-add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
-target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
-target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
+   target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
 
-add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
-target_link_libraries(test_batched_gemm_fp32 PRIVATE utility)
-target_link_libraries(test_batched_gemm_fp32 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
+   target_link_libraries(test_batched_gemm_fp32 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_fp32 PRIVATE device_batched_gemm_instance)
 
-add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
-target_link_libraries(test_batched_gemm_bf16 PRIVATE utility)
-target_link_libraries(test_batched_gemm_bf16 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
+   target_link_libraries(test_batched_gemm_bf16 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_bf16 PRIVATE device_batched_gemm_instance)
 
-add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
-target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
-target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
+   target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm/batched_gemm_bf16.cpp b/test/batched_gemm/batched_gemm_bf16.cpp
index 78be540627850426a075e2bde1516410ef070756..5d12a1e956d469d0a1be7301c993bfd981ebafc3 100644
--- a/test/batched_gemm/batched_gemm_bf16.cpp
+++ b/test/batched_gemm/batched_gemm_bf16.cpp
@@ -1,10 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = ck::bhalf_t;
 using BDataType = ck::bhalf_t;
@@ -12,6 +14,8 @@ using CDataType = ck::bhalf_t;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM bf16: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm/batched_gemm_fp16.cpp b/test/batched_gemm/batched_gemm_fp16.cpp
index 6cbbedf6774d2f90f85e5128c5bae8fbe912dc8e..a2b61d951a724bce18836c229ca7ef865336d849 100644
--- a/test/batched_gemm/batched_gemm_fp16.cpp
+++ b/test/batched_gemm/batched_gemm_fp16.cpp
@@ -1,10 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = ck::half_t;
 using BDataType = ck::half_t;
@@ -12,6 +14,8 @@ using CDataType = ck::half_t;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM fp16: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm/batched_gemm_fp32.cpp b/test/batched_gemm/batched_gemm_fp32.cpp
index c9e565e264b493d6f7f1b248fe6d307150a4c989..2b18d166e68f96e53670b4783038228b22b41cec 100644
--- a/test/batched_gemm/batched_gemm_fp32.cpp
+++ b/test/batched_gemm/batched_gemm_fp32.cpp
@@ -1,10 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = float;
 using BDataType = float;
@@ -12,6 +14,8 @@ using CDataType = float;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM fp32: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm/batched_gemm_int8.cpp b/test/batched_gemm/batched_gemm_int8.cpp
index 4da941a5766bc6decc5adc938b4ce39da5d8a6db..f607eaa84b996a0c28394647021953c5da6017a0 100644
--- a/test/batched_gemm/batched_gemm_int8.cpp
+++ b/test/batched_gemm/batched_gemm_int8.cpp
@@ -1,10 +1,12 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = int8_t;
 using BDataType = int8_t;
@@ -12,6 +14,8 @@ using CDataType = int8_t;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM int8: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm_gemm/CMakeLists.txt b/test/batched_gemm_gemm/CMakeLists.txt
index 386809717f242c5a23dd3c49437b4ebf055243eb..858efd837980658f20e0929ad93bc97e92ba7909 100644
--- a/test/batched_gemm_gemm/CMakeLists.txt
+++ b/test/batched_gemm_gemm/CMakeLists.txt
@@ -1,5 +1,11 @@
-add_custom_target(test_batched_gemm_gemm)
-
-add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
-target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
-add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
\ No newline at end of file
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(test_batched_gemm_gemm)
+   add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
+   target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
+   add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp b/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp
index aa113de219437e900b2de8739ddaffb2caf3cefa..1a8d5c2e55906fb77cfc3fef7d6c985c37eefefb 100644
--- a/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp
+++ b/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_gemm_util.hpp"
diff --git a/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp b/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp
index 53c4d37c44781be976f917362d0c433d1ec1711f..b0fffc466efd7de5f9a8055cb3f7e372bfad2ef7 100644
--- a/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp
+++ b/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_multi_d/CMakeLists.txt b/test/batched_gemm_multi_d/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..45a306551f28295a33ac0eab54cdcd04052942cc
--- /dev/null
+++ b/test/batched_gemm_multi_d/CMakeLists.txt
@@ -0,0 +1,5 @@
+# TODO: Enable for gfx90a after complier fix
+if(NOT GPU_TARGETS MATCHES "gfx90a")
+    add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
+    target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
+endif()
diff --git a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..6cf6145cd77fd1a39e1475b7e2a7a9100f83508c
--- /dev/null
+++ b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <gtest/gtest.h>
+
+#include "profiler/profile_batched_gemm_impl.hpp"
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp"
+
+namespace {
+using F16 = ck::half_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+template <typename Tuple>
+class TestBatchedGemmMultiD : public ::testing::Test
+{
+    protected:
+    using ALayout = std::tuple_element_t<0, Tuple>;
+    using BLayout = std::tuple_element_t<1, Tuple>;
+    using CLayout = std::tuple_element_t<2, Tuple>;
+
+    static constexpr int M          = 512;
+    static constexpr int N          = 256;
+    static constexpr int K          = 128;
+    static constexpr int BatchCount = 3;
+
+    template <typename DataType>
+    void Run()
+    {
+        using namespace ck::tensor_operation::device;
+
+        const bool pass =
+            ck::profiler::profile_batched_gemm_impl<DataType,
+                                                    DataType,
+                                                    DataType,
+                                                    ALayout,
+                                                    BLayout,
+                                                    CLayout,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    DeviceBatchedGemmMultiD<ALayout,
+                                                                            BLayout,
+                                                                            Empty_Tuple,
+                                                                            CLayout,
+                                                                            DataType,
+                                                                            DataType,
+                                                                            Empty_Tuple,
+                                                                            DataType,
+                                                                            PassThrough,
+                                                                            PassThrough,
+                                                                            PassThrough>>(
+                true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+        EXPECT_TRUE(pass);
+    }
+};
+
+using KernelTypes = ::testing::Types<std::tuple<Row, Row, Row>,
+                                     std::tuple<Row, Col, Row>,
+                                     std::tuple<Col, Row, Row>,
+                                     std::tuple<Col, Col, Row>>;
+} // namespace
+
+TYPED_TEST_SUITE(TestBatchedGemmMultiD, KernelTypes);
+
+TYPED_TEST(TestBatchedGemmMultiD, f16) { this->template Run<F16>(); }
+
+TYPED_TEST(TestBatchedGemmMultiD, int8) { this->template Run<int8_t>(); }
diff --git a/test/batched_gemm_reduce/CMakeLists.txt b/test/batched_gemm_reduce/CMakeLists.txt
index 4dc0b0825744294d219a39d46ec4f8aec06fdf93..0710f4647a7a8fe1b346ed7fedfa0bd3e3d41354 100644
--- a/test/batched_gemm_reduce/CMakeLists.txt
+++ b/test/batched_gemm_reduce/CMakeLists.txt
@@ -1,3 +1,10 @@
-add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
-target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
-target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
+   target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp b/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp
index b150ce50d166d5a673c6b34c468b7988f3b4d91f..dd2638ce899bee16accfcfb85c402909b2c2a05c 100644
--- a/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp
+++ b/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_softmax_gemm/CMakeLists.txt b/test/batched_gemm_softmax_gemm/CMakeLists.txt
index 1ceecefb5f2c0a86e6d951948c84a2286023773f..984cc3c160d04ee40cdb227dfe339b5959a539f9 100644
--- a/test/batched_gemm_softmax_gemm/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm/CMakeLists.txt
@@ -1,5 +1,11 @@
-add_custom_target(test_batched_gemm_softmax_gemm)
-
-add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
-target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
-add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
\ No newline at end of file
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(test_batched_gemm_softmax_gemm)
+   add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
+   target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
+   add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp
index 5df7769d5f675e906e350738751923d26b6af39d..cb46a995c61c4dc15c9fcbd9c2ca07b850f90069 100644
--- a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp
+++ b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_softmax_gemm_util.hpp"
diff --git a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp
index 98debe19c3c0b419f46dd08321f43f846536772a..d8ee744c6098843e9e78e21ac59a5f3733f40aa8 100644
--- a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp
+++ b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index f858d9f2087632ad3309ccc5cf7a3e851bca6c4b..4d04be3faa5af1adb2ae37793022ab794a27d9c3 100644
--- a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -1,8 +1,22 @@
-add_custom_target(test_batched_gemm_softmax_gemm_permute)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(test_batched_gemm_softmax_gemm_permute)
 
-add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
-add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
-target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
-add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
\ No newline at end of file
+   add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
+   add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
+   target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
+
+   add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
+   add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
+   target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ef88ce6d81845d9ddb61fc00159899f0d11daa5f
--- /dev/null
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp
@@ -0,0 +1,182 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "test_batched_gemm_bias_softmax_gemm_permute_util.hpp"
+
+template <typename Tuple>
+class TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16
+    : public TestBatchedGemmMaskingScaleSoftmaxGemmPermute<Tuple>
+{
+};
+
+using I1_t = ck::Number<1>;
+using I2_t = ck::Number<2>;
+
+using MaskDisabled_t =
+    ck::integral_constant<MaskingSpecialization, MaskingSpecialization::MaskDisabled>;
+using MaskOutUpperTriangle_t =
+    ck::integral_constant<MaskingSpecialization, MaskingSpecialization::MaskOutUpperTriangle>;
+
+// clang-format off
+using KernelTypes = ::testing::Types<
+    std::tuple<I2_t, I1_t, I1_t, I1_t, I1_t, BF16, BF16, BF16, BF16, ck::Tuple<BF16>, ck::Tuple<>, MaskDisabled_t>,
+    std::tuple<I2_t, I1_t, I1_t, I1_t, I1_t, BF16, BF16, BF16, BF16, ck::Tuple<BF16>, ck::Tuple<>, MaskOutUpperTriangle_t>
+    >;
+// clang-format on
+
+TYPED_TEST_SUITE(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, KernelTypes);
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, DISABLED_Test_BF16) { this->Run(); }
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_PadM)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {136, 128, 32, 128, 2, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_PadN)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 136, 32, 128, 3, 2},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_PadK)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 40, 128, 2, 4},
+        {128, 128, 136, 128, 4, 2},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_PadO)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 32, 136, 1, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_OddM)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {129, 128, 32, 128, 2, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_OddN)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 129, 32, 128, 4, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_OddK)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 33, 128, 2, 3},
+        {128, 128, 129, 128, 2, 3},
+    };
+    this->Run();
+}
+
+// If kernel B1Layout is RowMajor, expect not to support odd O size
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_OddO)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 32, 129, 2, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, DISABLED_Bench_BF16_IrregularK)
+{
+    this->lengths_ = std::vector<std::vector<int>>{{256, 256, 160, 160, 1, 16},
+                                                   {256, 64, 160, 64, 1, 16},
+                                                   {1024, 1024, 80, 80, 1, 16},
+                                                   {1024, 64, 80, 64, 1, 16},
+                                                   {4096, 4096, 40, 40, 1, 16},
+                                                   {4096, 64, 40, 64, 1, 16}};
+    this->bench_   = true;
+    this->verify_  = false;
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, DISABLED_Bench_BF16)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {256, 256, 64, 64, 48, 16},
+        {256, 256, 128, 128, 48, 16},
+        {512, 512, 64, 64, 48, 16},
+        {512, 512, 128, 128, 48, 16},
+        {1024, 1024, 64, 64, 48, 16},
+        {1024, 1024, 128, 128, 48, 16},
+        {2048, 2048, 64, 64, 48, 16},
+        {2048, 2048, 128, 128, 48, 16},
+        {4096, 4096, 64, 64, 48, 16},
+        {4096, 4096, 128, 128, 48, 16},
+    };
+    this->bench_  = true;
+    this->verify_ = false;
+    this->Run();
+}
+
+using ck::tensor_operation::device::GemmSpecialization;
+
+TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteInterface, GemmSpecializationSizeMatch)
+{
+    int P = 120; // requires padding
+    int Q = 128; // do not require padding
+
+    // IsSupported(M, N, K, O)
+    // clang-format off
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::Default>{}.IsSupported(Q, Q, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MPadding>{}.IsSupported(P, Q, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::NPadding>{}.IsSupported(Q, P, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::KPadding>{}.IsSupported(Q, Q, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNPadding>{}.IsSupported(P, P, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MKPadding>{}.IsSupported(P, Q, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::NKPadding>{}.IsSupported(Q, P, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNKPadding>{}.IsSupported(P, P, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::OPadding>{}.IsSupported(Q, Q, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MOPadding>{}.IsSupported(P, Q, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::NOPadding>{}.IsSupported(Q, P, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::KOPadding>{}.IsSupported(Q, Q, P, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNOPadding>{}.IsSupported(P, P, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MKOPadding>{}.IsSupported(P, Q, P, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::NKOPadding>{}.IsSupported(Q, P, P, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(P, P, P, P));
+    // clang-format on
+}
+
+TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteInterface, GemmSpecializationSizeMismatch)
+{
+    // IsSupported(M, N, K, O)
+    // clang-format off
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::Default>{}.IsSupported(128, 128, 120, 128));
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNKPadding>{}.IsSupported(128, 128, 128, 120));
+    // Kernel can't support odd K size because SrcVectorDim == KDim and must satisfy SizeKRaw % ABSrcScalarPerVector == 0
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(128, 128, 129, 128));
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(128, 128, 130, 128));
+    // Kernel can't support odd O size because SrcVectorDim == ODim and must satisfy SizeORaw % B1SrcScalarPerVector == 0
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(128, 128, 128, 129));
+    // clang-format on
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, AdhocTest)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {49, 49, 64, 64, 4, 6},
+        {64, 49, 64, 64, 4, 6},
+        {1020, 1020, 64, 128, 4, 6},
+        {576, 576, 64, 64, 4, 6},
+    };
+    this->Run();
+}
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b38b10d1953091bc7d0739a38fa184d3d51a94f1
--- /dev/null
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp
@@ -0,0 +1,182 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "test_batched_gemm_softmax_gemm_permute_util.hpp"
+
+template <typename Tuple>
+class TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16
+    : public TestBatchedGemmMaskingScaleSoftmaxGemmPermute<Tuple>
+{
+};
+
+using I1_t = ck::Number<1>;
+using I2_t = ck::Number<2>;
+
+using MaskDisabled_t =
+    ck::integral_constant<MaskingSpecialization, MaskingSpecialization::MaskDisabled>;
+using MaskOutUpperTriangle_t =
+    ck::integral_constant<MaskingSpecialization, MaskingSpecialization::MaskOutUpperTriangle>;
+
+// clang-format off
+using KernelTypes = ::testing::Types<
+    std::tuple<I2_t, I1_t, I1_t, I1_t, I1_t, F16, F16, F16, F16, ck::Tuple<F16>, ck::Tuple<>, MaskDisabled_t>,
+    std::tuple<I2_t, I1_t, I1_t, I1_t, I1_t, F16, F16, F16, F16, ck::Tuple<F16>, ck::Tuple<>, MaskOutUpperTriangle_t>
+    >;
+// clang-format on
+
+TYPED_TEST_SUITE(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, KernelTypes);
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16) { this->Run(); }
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_PadM)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {136, 128, 32, 128, 2, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_PadN)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 136, 32, 128, 3, 2},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_PadK)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 40, 128, 2, 4},
+        {128, 128, 136, 128, 4, 2},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_PadO)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 32, 136, 1, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_OddM)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {129, 128, 32, 128, 2, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_OddN)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 129, 32, 128, 4, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_OddK)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 33, 128, 2, 3},
+        {128, 128, 129, 128, 2, 3},
+    };
+    this->Run();
+}
+
+// If kernel B1Layout is RowMajor, expect not to support odd O size
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, Test_FP16_OddO)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {128, 128, 32, 129, 2, 3},
+    };
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, DISABLED_Bench_FP16_IrregularK)
+{
+    this->lengths_ = std::vector<std::vector<int>>{{256, 256, 160, 160, 1, 16},
+                                                   {256, 64, 160, 64, 1, 16},
+                                                   {1024, 1024, 80, 80, 1, 16},
+                                                   {1024, 64, 80, 64, 1, 16},
+                                                   {4096, 4096, 40, 40, 1, 16},
+                                                   {4096, 64, 40, 64, 1, 16}};
+    this->bench_   = true;
+    this->verify_  = false;
+    this->Run();
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, DISABLED_Bench_FP16)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {256, 256, 64, 64, 48, 16},
+        {256, 256, 128, 128, 48, 16},
+        {512, 512, 64, 64, 48, 16},
+        {512, 512, 128, 128, 48, 16},
+        {1024, 1024, 64, 64, 48, 16},
+        {1024, 1024, 128, 128, 48, 16},
+        {2048, 2048, 64, 64, 48, 16},
+        {2048, 2048, 128, 128, 48, 16},
+        {4096, 4096, 64, 64, 48, 16},
+        {4096, 4096, 128, 128, 48, 16},
+    };
+    this->bench_  = true;
+    this->verify_ = false;
+    this->Run();
+}
+
+using ck::tensor_operation::device::GemmSpecialization;
+
+TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteInterface, GemmSpecializationSizeMatch)
+{
+    int P = 120; // requires padding
+    int Q = 128; // do not require padding
+
+    // IsSupported(M, N, K, O)
+    // clang-format off
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::Default>{}.IsSupported(Q, Q, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MPadding>{}.IsSupported(P, Q, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::NPadding>{}.IsSupported(Q, P, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::KPadding>{}.IsSupported(Q, Q, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNPadding>{}.IsSupported(P, P, Q, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MKPadding>{}.IsSupported(P, Q, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::NKPadding>{}.IsSupported(Q, P, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNKPadding>{}.IsSupported(P, P, P, Q));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::OPadding>{}.IsSupported(Q, Q, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MOPadding>{}.IsSupported(P, Q, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::NOPadding>{}.IsSupported(Q, P, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::KOPadding>{}.IsSupported(Q, Q, P, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNOPadding>{}.IsSupported(P, P, Q, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MKOPadding>{}.IsSupported(P, Q, P, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::NKOPadding>{}.IsSupported(Q, P, P, P));
+    EXPECT_TRUE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(P, P, P, P));
+    // clang-format on
+}
+
+TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteInterface, GemmSpecializationSizeMismatch)
+{
+    // IsSupported(M, N, K, O)
+    // clang-format off
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::Default>{}.IsSupported(128, 128, 120, 128));
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNKPadding>{}.IsSupported(128, 128, 128, 120));
+    // Kernel can't support odd K size because SrcVectorDim == KDim and must satisfy SizeKRaw % ABSrcScalarPerVector == 0
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(128, 128, 129, 128));
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(128, 128, 130, 128));
+    // Kernel can't support odd O size because SrcVectorDim == ODim and must satisfy SizeORaw % B1SrcScalarPerVector == 0
+    EXPECT_FALSE(DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128<GemmSpecialization::MNKOPadding>{}.IsSupported(128, 128, 128, 129));
+    // clang-format on
+}
+
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteFP16, AdhocTest)
+{
+    this->lengths_ = std::vector<std::vector<int>>{
+        {49, 49, 64, 64, 4, 6},
+        {64, 49, 64, 64, 4, 6},
+        {1020, 1020, 64, 128, 4, 6},
+        {576, 576, 64, 64, 4, 6},
+    };
+    this->Run();
+}
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_util.hpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_util.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..d7c39367c8a19290718b315251581f4f3f626add
--- /dev/null
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_util.hpp
@@ -0,0 +1,380 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include <vector>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp"
+
+using ck::tensor_operation::device::GemmSpecialization;
+using ck::tensor_operation::device::MaskingSpecialization;
+using ck::tensor_operation::device::TensorSpecialization;
+
+template <ck::index_t N>
+using I = ck::Number<N>;
+
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <typename Tuple>
+struct TestBatchedGemmMaskingScaleSoftmaxGemmPermute : public ::testing::Test
+{
+    using NumDimGType      = std::tuple_element_t<0, Tuple>;
+    using NumDimMType      = std::tuple_element_t<1, Tuple>;
+    using NumDimNType      = std::tuple_element_t<2, Tuple>;
+    using NumDimKType      = std::tuple_element_t<3, Tuple>;
+    using NumDimOType      = std::tuple_element_t<4, Tuple>;
+    using ADataType        = std::tuple_element_t<5, Tuple>;
+    using B0DataType       = std::tuple_element_t<6, Tuple>;
+    using B1DataType       = std::tuple_element_t<7, Tuple>;
+    using CDataType        = std::tuple_element_t<8, Tuple>;
+    using Acc0BiasDataType = std::tuple_element_t<9, Tuple>;
+    using Acc1BiasDataType = std::tuple_element_t<10, Tuple>;
+    using MaskingType      = std::tuple_element_t<11, Tuple>;
+
+    std::vector<std::vector<int>> lengths_ = {
+        {256, 256, 64, 64, 6, 4},
+        {256, 256, 128, 128, 4, 6},
+        {512, 512, 64, 64, 3, 2},
+        {512, 512, 128, 128, 2, 3},
+        {1024, 1024, 64, 64, 3, 1},
+        {1024, 1024, 128, 128, 1, 1},
+    };
+    bool bench_  = false;
+    bool verify_ = true;
+
+    void RunSingle(int M, int N, int K, int O, int G0, int G1)
+    {
+        bool pass =
+            ck::profiler::profile_batched_gemm_bias_softmax_gemm_permute_impl<NumDimGType::value,
+                                                                              NumDimMType::value,
+                                                                              NumDimNType::value,
+                                                                              NumDimKType::value,
+                                                                              NumDimOType::value,
+                                                                              ADataType,
+                                                                              B0DataType,
+                                                                              B1DataType,
+                                                                              CDataType,
+                                                                              Acc0BiasDataType,
+                                                                              Acc1BiasDataType,
+                                                                              MaskingType::value>(
+                verify_, 2, false, bench_, M, N, K, O, G0, G1);
+
+        EXPECT_TRUE(pass);
+    }
+
+    void Run()
+    {
+        for(auto lengths : this->lengths_)
+        {
+            int M  = lengths[0];
+            int N  = lengths[1];
+            int K  = lengths[2];
+            int O  = lengths[3];
+            int G0 = lengths[4];
+            int G1 = lengths[5];
+
+            this->RunSingle(M, N, K, O, G0, G1);
+        }
+    }
+};
+
+template <GemmSpecialization GemmSpec>
+struct DeviceInstanceWrapper_G2M1N1K1O1_TNTT_FP16_M128_N128_K32_O128
+{
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+    using ScaleAdd    = ck::tensor_operation::element_wise::ScaleAdd;
+
+    template <ck::index_t... Is>
+    using S = ck::Sequence<Is...>;
+
+    using ADataType        = F16;
+    using B0DataType       = F16;
+    using B1DataType       = F16;
+    using AccDataType      = float;
+    using CShuffleDataType = F16;
+    using CDataType        = F16;
+
+    using AElementOp    = PassThrough;
+    using B0ElementOp   = PassThrough;
+    using Acc0ElementOp = ScaleAdd;
+    using B1ElementOp   = PassThrough;
+    using CElementOp    = PassThrough;
+
+    // static constexpr auto GemmSpec = std::tuple_element_t<0, Tuple>::value;
+
+    using DeviceGemmGemmInstance =
+        ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<
+            2,
+            1,
+            1,
+            1,
+            1,
+            ADataType,
+            B0DataType,
+            B1DataType,
+            CDataType,
+            ck::Tuple<F16>,
+            ck::Tuple<>,
+            AccDataType,
+            CShuffleDataType,
+            AElementOp,
+            B0ElementOp,
+            Acc0ElementOp,
+            B1ElementOp,
+            CElementOp,
+            GemmSpec,
+            TensorSpecialization::Default, // ATensorSpec
+            TensorSpecialization::Default, // B0TensorSpec
+            TensorSpecialization::Default, // B1TensorSpec
+            TensorSpecialization::Default, // CTensorSpec
+            1,
+            256,
+            128,         // MPerBlock
+            128,         // NPerBlock
+            32,          // KPerBlock
+            128,         // Gemm1NPerBlock
+            32,          // Gemm1KPerBlock
+            8,           // AK1
+            8,           // BK1
+            2,           // B1K1
+            32,          // MPerXDL
+            32,          // NPerXDL
+            1,           // MXdlPerWave
+            4,           // NXdlPerWave
+            4,           // Gemm1NXdlPerWave
+            S<4, 64, 1>, // ABlockTransfer
+            S<1, 0, 2>,
+            S<1, 0, 2>,
+            2,
+            8,
+            8,
+            true,
+            S<4, 64, 1>, // BBlockTransfer
+            S<1, 0, 2>,
+            S<1, 0, 2>,
+            2,
+            8,
+            8,
+            true,
+            S<8, 32, 1>, // B1BlockTransfer
+            S<0, 2, 1>,
+            S<0, 2, 1>,
+            1,
+            4,
+            2,
+            false,
+            1,              // CShuffleMXdlPerWavePerShuffle
+            2,              // CShuffleNXdlPerWavePerShuffle
+            S<1, 32, 1, 8>, // CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
+            8,              // CShuffleBlockTransferScalarPerVector_NPerBlock
+            MaskingSpecialization::MaskOutUpperTriangle>; // MaskOutUpperTriangle
+
+    bool IsSupported(int M, int N, int K, int O)
+    {
+        const int G0 = 1, G1 = 1;
+
+        // A layout [G0, M, G1, K]
+        std::vector<ck::index_t> a_gs_ms_ks_lengths{G0, G1, M, K};
+        std::vector<ck::index_t> a_gs_ms_ks_strides{M * G1 * K, K, G1 * K, 1};
+
+        // B0 layout [G0, N, G1, K]
+        std::vector<ck::index_t> b0_gs_ns_ks_lengths{G0, G1, N, K};
+        std::vector<ck::index_t> b0_gs_ns_ks_strides{N * G1 * K, K, G1 * K, 1};
+
+        // B1 layout [G0, N, G1, O]
+        std::vector<ck::index_t> b1_gs_os_ns_lengths{G0, G1, O, N};
+        std::vector<ck::index_t> b1_gs_os_ns_strides{N * G1 * O, O, 1, G1 * O};
+
+        // C layout [G0, M, G1, O]
+        std::vector<ck::index_t> c_gs_ms_os_lengths{G0, G1, M, O};
+        std::vector<ck::index_t> c_gs_ms_os_strides{M * G1 * O, O, G1 * O, 1};
+
+        // D layout [G0, M, G1, N]
+        std::vector<ck::index_t> d0_gs_ms_ns_lengths{G0, G1, M, N};
+        std::vector<ck::index_t> d0_gs_ms_ns_strides{M * G1 * N, N, G1 * N, 1};
+
+        auto gemm     = DeviceGemmGemmInstance{};
+        auto invoker  = gemm.MakeInvoker();
+        auto argument = gemm.MakeArgument(static_cast<ADataType*>(nullptr),
+                                          static_cast<B0DataType*>(nullptr),
+                                          static_cast<B1DataType*>(nullptr),
+                                          static_cast<CDataType*>(nullptr),
+                                          std::array<void*, 1>{nullptr}, // p_acc0_biases
+                                          {},                            // p_acc1_biases
+                                          a_gs_ms_ks_lengths,
+                                          a_gs_ms_ks_strides,
+                                          b0_gs_ns_ks_lengths,
+                                          b0_gs_ns_ks_strides,
+                                          b1_gs_os_ns_lengths,
+                                          b1_gs_os_ns_strides,
+                                          c_gs_ms_os_lengths,
+                                          c_gs_ms_os_strides,
+                                          std::array<std::vector<ck::index_t>, 1>{
+                                              d0_gs_ms_ns_lengths}, // acc0_biases_gs_ms_ns_lengths
+                                          std::array<std::vector<ck::index_t>, 1>{
+                                              d0_gs_ms_ns_strides}, // acc0_biases_gs_ms_ns_strides
+                                          {},                       // acc1_biases_gs_ms_os_lengths
+                                          {},                       // acc1_biases_gs_ms_os_strides
+                                          PassThrough{},            // a_element_op
+                                          PassThrough{},            // b0_element_op
+                                          Acc0ElementOp{1.f},       // acc0_element_op
+                                          PassThrough{},            // b1_element_op
+                                          PassThrough{});           // c_element_op
+
+        return gemm.IsSupportedArgument(argument);
+    }
+};
+
+template <GemmSpecialization GemmSpec>
+struct DeviceInstanceWrapper_G2M1N1K1O1_TNTT_BF16_M128_N128_K32_O128
+{
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+    using ScaleAdd    = ck::tensor_operation::element_wise::ScaleAdd;
+
+    template <ck::index_t... Is>
+    using S = ck::Sequence<Is...>;
+
+    using ADataType        = BF16;
+    using B0DataType       = BF16;
+    using B1DataType       = BF16;
+    using AccDataType      = float;
+    using CShuffleDataType = BF16;
+    using CDataType        = BF16;
+
+    using AElementOp    = PassThrough;
+    using B0ElementOp   = PassThrough;
+    using Acc0ElementOp = ScaleAdd;
+    using B1ElementOp   = PassThrough;
+    using CElementOp    = PassThrough;
+
+    // static constexpr auto GemmSpec = std::tuple_element_t<0, Tuple>::value;
+
+    using DeviceGemmGemmInstance =
+        ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<
+            2,
+            1,
+            1,
+            1,
+            1,
+            ADataType,
+            B0DataType,
+            B1DataType,
+            CDataType,
+            ck::Tuple<BF16>,
+            ck::Tuple<>,
+            AccDataType,
+            CShuffleDataType,
+            AElementOp,
+            B0ElementOp,
+            Acc0ElementOp,
+            B1ElementOp,
+            CElementOp,
+            GemmSpec,
+            TensorSpecialization::Default, // ATensorSpec
+            TensorSpecialization::Default, // B0TensorSpec
+            TensorSpecialization::Default, // B1TensorSpec
+            TensorSpecialization::Default, // CTensorSpec
+            1,
+            256,
+            128,         // MPerBlock
+            128,         // NPerBlock
+            32,          // KPerBlock
+            128,         // Gemm1NPerBlock
+            32,          // Gemm1KPerBlock
+            8,           // AK1
+            8,           // BK1
+            2,           // B1K1
+            32,          // MPerXDL
+            32,          // NPerXDL
+            1,           // MXdlPerWave
+            4,           // NXdlPerWave
+            4,           // Gemm1NXdlPerWave
+            S<4, 64, 1>, // ABlockTransfer
+            S<1, 0, 2>,
+            S<1, 0, 2>,
+            2,
+            8,
+            8,
+            true,
+            S<4, 64, 1>, // BBlockTransfer
+            S<1, 0, 2>,
+            S<1, 0, 2>,
+            2,
+            8,
+            8,
+            true,
+            S<8, 32, 1>, // B1BlockTransfer
+            S<0, 2, 1>,
+            S<0, 2, 1>,
+            1,
+            4,
+            2,
+            false,
+            1,              // CShuffleMXdlPerWavePerShuffle
+            2,              // CShuffleNXdlPerWavePerShuffle
+            S<1, 32, 1, 8>, // CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
+            8,              // CShuffleBlockTransferScalarPerVector_NPerBlock
+            MaskingSpecialization::MaskOutUpperTriangle>; // MaskOutUpperTriangle
+
+    bool IsSupported(int M, int N, int K, int O)
+    {
+        const int G0 = 1, G1 = 1;
+
+        // A layout [G0, M, G1, K]
+        std::vector<ck::index_t> a_gs_ms_ks_lengths{G0, G1, M, K};
+        std::vector<ck::index_t> a_gs_ms_ks_strides{M * G1 * K, K, G1 * K, 1};
+
+        // B0 layout [G0, N, G1, K]
+        std::vector<ck::index_t> b0_gs_ns_ks_lengths{G0, G1, N, K};
+        std::vector<ck::index_t> b0_gs_ns_ks_strides{N * G1 * K, K, G1 * K, 1};
+
+        // B1 layout [G0, N, G1, O]
+        std::vector<ck::index_t> b1_gs_os_ns_lengths{G0, G1, O, N};
+        std::vector<ck::index_t> b1_gs_os_ns_strides{N * G1 * O, O, 1, G1 * O};
+
+        // C layout [G0, M, G1, O]
+        std::vector<ck::index_t> c_gs_ms_os_lengths{G0, G1, M, O};
+        std::vector<ck::index_t> c_gs_ms_os_strides{M * G1 * O, O, G1 * O, 1};
+
+        // D layout [G0, M, G1, N]
+        std::vector<ck::index_t> d0_gs_ms_ns_lengths{G0, G1, M, N};
+        std::vector<ck::index_t> d0_gs_ms_ns_strides{M * G1 * N, N, G1 * N, 1};
+
+        auto gemm     = DeviceGemmGemmInstance{};
+        auto invoker  = gemm.MakeInvoker();
+        auto argument = gemm.MakeArgument(static_cast<ADataType*>(nullptr),
+                                          static_cast<B0DataType*>(nullptr),
+                                          static_cast<B1DataType*>(nullptr),
+                                          static_cast<CDataType*>(nullptr),
+                                          std::array<void*, 1>{nullptr}, // p_acc0_biases
+                                          {},                            // p_acc1_biases
+                                          a_gs_ms_ks_lengths,
+                                          a_gs_ms_ks_strides,
+                                          b0_gs_ns_ks_lengths,
+                                          b0_gs_ns_ks_strides,
+                                          b1_gs_os_ns_lengths,
+                                          b1_gs_os_ns_strides,
+                                          c_gs_ms_os_lengths,
+                                          c_gs_ms_os_strides,
+                                          std::array<std::vector<ck::index_t>, 1>{
+                                              d0_gs_ms_ns_lengths}, // acc0_biases_gs_ms_ns_lengths
+                                          std::array<std::vector<ck::index_t>, 1>{
+                                              d0_gs_ms_ns_strides}, // acc0_biases_gs_ms_ns_strides
+                                          {},                       // acc1_biases_gs_ms_os_lengths
+                                          {},                       // acc1_biases_gs_ms_os_strides
+                                          PassThrough{},            // a_element_op
+                                          PassThrough{},            // b0_element_op
+                                          Acc0ElementOp{1.f},       // acc0_element_op
+                                          PassThrough{},            // b1_element_op
+                                          PassThrough{});           // c_element_op
+
+        return gemm.IsSupportedArgument(argument);
+    }
+};
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp
index 4a775e6d928d646e99887eac3dd174fee8450f84..8e0baede113943c0bcb7d6f929217b2be94f0073 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_softmax_gemm_permute_util.hpp"
@@ -27,7 +27,7 @@ using KernelTypes = ::testing::Types<
 
 TYPED_TEST_SUITE(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, KernelTypes);
 
-TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, DISABLED_Test_BF16) { this->Run(); }
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16) { this->Run(); }
 
 TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_PadM)
 {
@@ -96,7 +96,7 @@ TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Test_BF16_OddO)
     this->Run();
 }
 
-TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, DISABLED_Bench_BF16_IrregularK)
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Bench_BF16_IrregularK)
 {
     this->lengths_ = std::vector<std::vector<int>>{{256, 256, 160, 160, 1, 16},
                                                    {256, 64, 160, 64, 1, 16},
@@ -109,7 +109,7 @@ TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, DISABLED_Bench_BF1
     this->Run();
 }
 
-TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, DISABLED_Bench_BF16)
+TYPED_TEST(TestBatchedGemmMaskingScaleSoftmaxGemmPermuteBF16, Bench_BF16)
 {
     this->lengths_ = std::vector<std::vector<int>>{
         {256, 256, 64, 64, 48, 16},
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp
index 293acd60155ace98eaea89a11eb03fa6710fd877..81d404109fddc4321d870d1c3b8c0a2fe309c9c4 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_softmax_gemm_permute_util.hpp"
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp
index 912bbc91edd51d06c827fc31495d2955bea85a2b..9df03ffd2a11211600872957e9686e105bf8b2f2 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batchnorm/CMakeLists.txt b/test/batchnorm/CMakeLists.txt
index 52f15086822cfed89a89a3850e2ad966b2e2d762..2a528f9c37fec08124b96bab9aa57464b379ff90 100644
--- a/test/batchnorm/CMakeLists.txt
+++ b/test/batchnorm/CMakeLists.txt
@@ -1,4 +1,6 @@
 add_gtest_executable(test_batchnorm_fwd_rank_4 batchnorm_fwd_rank_4.cpp)
 add_gtest_executable(test_batchnorm_bwd_rank_4 batchnorm_bwd_rank_4.cpp)
+add_gtest_executable(test_batchnorm_infer_rank_4 batchnorm_infer_rank_4.cpp)
 target_link_libraries(test_batchnorm_fwd_rank_4 PRIVATE utility device_batchnorm_instance)
 target_link_libraries(test_batchnorm_bwd_rank_4 PRIVATE utility device_batchnorm_instance)
+target_link_libraries(test_batchnorm_infer_rank_4 PRIVATE utility device_batchnorm_instance)
diff --git a/test/batchnorm/batchnorm_bwd_rank_4.cpp b/test/batchnorm/batchnorm_bwd_rank_4.cpp
index caa7331ea2c3e6dfe57a41272c5f4fba56662e39..a4696cf2a39a774c86d4a68f1a0edf0ce31a3f58 100644
--- a/test/batchnorm/batchnorm_bwd_rank_4.cpp
+++ b/test/batchnorm/batchnorm_bwd_rank_4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/batchnorm/batchnorm_fwd_rank_4.cpp b/test/batchnorm/batchnorm_fwd_rank_4.cpp
index 13aef7d6bfc9f222e663b951a2f9d44e81fc93bc..9b6fbd0f662648426aad3f9c316604510ff6e1b1 100644
--- a/test/batchnorm/batchnorm_fwd_rank_4.cpp
+++ b/test/batchnorm/batchnorm_fwd_rank_4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/batchnorm/batchnorm_infer_rank_4.cpp b/test/batchnorm/batchnorm_infer_rank_4.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ecb4043b361b058ea82b91d98232eadb694ead33
--- /dev/null
+++ b/test/batchnorm/batchnorm_infer_rank_4.cpp
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <vector>
+#include <tuple>
+#include <gtest/gtest.h>
+
+#include "profiler/profile_batchnorm_infer_impl.hpp"
+
+using F16  = ck::half_t;
+using F32  = float;
+using BF16 = ck::bhalf_t;
+using F64  = double;
+
+template <typename Tuple>
+class TestBatchNormInferRank4 : public ::testing::Test
+{
+    private:
+    const double epsilon = std::numeric_limits<float>::epsilon();
+
+    protected:
+    using XDataType       = std::tuple_element_t<0, Tuple>;
+    using YDataType       = std::tuple_element_t<1, Tuple>;
+    using AccDataType     = std::tuple_element_t<2, Tuple>;
+    using ScaleDataType   = std::tuple_element_t<3, Tuple>;
+    using BiasDataType    = std::tuple_element_t<4, Tuple>;
+    using MeanVarDataType = std::tuple_element_t<5, Tuple>;
+
+    std::vector<std::vector<size_t>> list_of_lengths = {
+        {128, 16, 3, 1024}, {128, 16, 6, 512}, {4, 4, 4, 4}, {32, 32, 32, 32}};
+    std::vector<int> reduceDims;
+
+    template <int NumReduceDim>
+    void Run()
+    {
+        for(auto& inOutLengths : list_of_lengths)
+        {
+            bool pass = true;
+
+            EXPECT_FALSE(reduceDims.size() != NumReduceDim);
+
+            pass = pass && ck::profiler::profile_batchnorm_infer_impl<XDataType,
+                                                                      YDataType,
+                                                                      AccDataType,
+                                                                      ScaleDataType,
+                                                                      BiasDataType,
+                                                                      MeanVarDataType,
+                                                                      4,
+                                                                      NumReduceDim>(
+                               true, 3, false, false, inOutLengths, reduceDims, epsilon);
+
+            pass = pass && ck::profiler::profile_batchnorm_infer_impl<XDataType,
+                                                                      YDataType,
+                                                                      AccDataType,
+                                                                      ScaleDataType,
+                                                                      BiasDataType,
+                                                                      MeanVarDataType,
+                                                                      4,
+                                                                      NumReduceDim>(
+                               true, 3, false, false, inOutLengths, reduceDims, epsilon);
+
+            EXPECT_TRUE(pass);
+        }
+    }
+};
+
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, F32>,
+                                     std::tuple<F32, F32, F32, F32, F32, F32>,
+                                     std::tuple<BF16, BF16, F32, BF16, BF16, F32>,
+                                     std::tuple<F64, F64, F64, F64, F64, F64>>;
+
+TYPED_TEST_SUITE(TestBatchNormInferRank4, KernelTypes);
+
+// nhwc
+TYPED_TEST(TestBatchNormInferRank4, nhwc)
+{
+    this->reduceDims = {0, 1, 2};
+    this->template Run<3>();
+}
+
+// nchw
+TYPED_TEST(TestBatchNormInferRank4, nchw)
+{
+    this->reduceDims = {0, 2, 3};
+    this->template Run<3>();
+}
diff --git a/test/block_to_ctile_map/test_block_to_ctile_map.cpp b/test/block_to_ctile_map/test_block_to_ctile_map.cpp
index 55d9b59f489203cf68b5aebc1c5d9642801435c0..b8e349eda16f53d97b9655840f9803ef8e6a7ace 100644
--- a/test/block_to_ctile_map/test_block_to_ctile_map.cpp
+++ b/test/block_to_ctile_map/test_block_to_ctile_map.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/test/contraction/CMakeLists.txt b/test/contraction/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1f6e0ed34187f1cd66d7bc8a2e8f1d5a16982d42
--- /dev/null
+++ b/test/contraction/CMakeLists.txt
@@ -0,0 +1,11 @@
+add_gtest_executable(test_contraction test_contraction.cpp)
+target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+    add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
+    target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/contraction/test_contraction.cpp b/test/contraction/test_contraction.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c86b849235f8c0b9eacc8ec9e3cf22e965ce60cc
--- /dev/null
+++ b/test/contraction/test_contraction.cpp
@@ -0,0 +1,138 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <memory>
+#include <initializer_list>
+#include <vector>
+#include <tuple>
+#include <gtest/gtest.h>
+
+#include "profiler/profile_contraction_impl.hpp"
+
+using F32 = float;
+using F64 = double;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+using Scale    = ck::tensor_operation::element_wise::Scale;
+
+struct MemoryParams
+{
+    std::vector<ck::index_t> M;
+    std::vector<ck::index_t> N;
+    std::vector<ck::index_t> K;
+    std::vector<ck::index_t> StridesA;
+    std::vector<ck::index_t> StridesB;
+    std::vector<ck::index_t> StridesC;
+    std::vector<ck::index_t> StridesD;
+};
+
+template <typename Tuple>
+class TestContraction : public ::testing::Test
+{
+    protected:
+    using ALayout        = std::tuple_element_t<0, Tuple>;
+    using BLayout        = std::tuple_element_t<1, Tuple>;
+    using CDLayout       = std::tuple_element_t<2, Tuple>;
+    using DataType       = std::tuple_element_t<3, Tuple>;
+    using DTupleDataType = std::tuple_element_t<4, Tuple>;
+    using CDElementOp    = std::tuple_element_t<5, Tuple>;
+
+    std::vector<MemoryParams> list_of_memory_params = {{{32, 32},
+                                                        {32, 32},
+                                                        {32, 32},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1}},
+                                                       {{16, 16},
+                                                        {32, 32},
+                                                        {16, 16},
+                                                        {4096, 256, 16, 1},
+                                                        {16, 1, 8192, 256},
+                                                        {16384, 1024, 32, 1},
+                                                        {16384, 1024, 32, 1}}};
+
+    std::vector<ck::index_t> init_methods = {0, 1, 2};
+    std::unique_ptr<CDElementOp> p_cd_element_op;
+    void Run()
+    {
+        for(auto& memory_params : list_of_memory_params)
+        {
+            for(const ck::index_t init_method : init_methods)
+            {
+                bool pass =
+                    ck::profiler::profile_contraction_impl<ALayout,
+                                                           BLayout,
+                                                           CDLayout,
+                                                           DataType,
+                                                           DTupleDataType,
+                                                           CDElementOp>(true /*do_verification*/,
+                                                                        init_method,
+                                                                        false /*do_logs*/,
+                                                                        false /*time_kernel*/,
+                                                                        *p_cd_element_op,
+                                                                        memory_params.M,
+                                                                        memory_params.N,
+                                                                        memory_params.K,
+                                                                        memory_params.StridesA,
+                                                                        memory_params.StridesB,
+                                                                        memory_params.StridesC,
+                                                                        memory_params.StridesD);
+                EXPECT_TRUE(pass);
+            }
+        }
+    }
+};
+
+template <typename Tuple>
+class TestContractionScale : public TestContraction<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestContractionBilinear : public TestContraction<Tuple>
+{
+};
+
+using BilinearKernelTypes =
+    ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Col, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Col, Row, F64, ck::Tuple<F32>, Bilinear>>;
+
+using ScaleKernelTypes = ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Col, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Row, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Col, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Row, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Col, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Row, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Col, Row, F64, ck::Tuple<>, Scale>>;
+
+TYPED_TEST_SUITE(TestContractionBilinear, BilinearKernelTypes);
+TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes);
+
+TYPED_TEST(TestContractionBilinear, bilinear)
+{
+    this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
+    this->Run();
+    this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
+    this->Run();
+}
+
+TYPED_TEST(TestContractionScale, scale)
+{
+    this->p_cd_element_op = std::make_unique<Scale>(1.f);
+    this->Run();
+    this->p_cd_element_op = std::make_unique<Scale>(0.5f);
+    this->Run();
+}
diff --git a/test/contraction/test_contraction_interface.cpp b/test/contraction/test_contraction_interface.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..c9e720c597b1ca59e8924b250b3c021f51a17a80
--- /dev/null
+++ b/test/contraction/test_contraction_interface.cpp
@@ -0,0 +1,195 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <stdexcept>
+#include <vector>
+
+#include "gtest/gtest.h"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+
+using Pass     = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F32 = float;
+using F64 = double;
+
+template <ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t CDEBlockTransferScalarPerVector>
+class ContractionInstanceWrapper
+{
+    public:
+    static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+    static constexpr ck::index_t NumDim = 2;
+    // clang-format off
+    using ContractionDeviceInstance = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|         DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer|             ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer|              BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|                  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|           Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|                 ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |               |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|                   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |               |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |                           |               |               |          |                |               |               |                           |               |               |          |            |            |                             |                                |
+        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,         Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
+    // clang-format on
+
+    bool isSupported(std::vector<ck::index_t>& ADims,
+                     std::vector<ck::index_t>& BDims,
+                     std::vector<ck::index_t>& DDims,
+                     std::vector<ck::index_t>& EDims,
+                     std::vector<ck::index_t>& AStrides,
+                     std::vector<ck::index_t>& BStrides,
+                     std::vector<ck::index_t>& DStrides,
+                     std::vector<ck::index_t>& EStrides) const
+    {
+        auto contraction = ContractionDeviceInstance{};
+
+        auto argument = contraction.MakeArgument(nullptr,
+                                                 nullptr,
+                                                 std::array<const void*, 1>{nullptr},
+                                                 nullptr,
+                                                 ADims,
+                                                 AStrides,
+                                                 BDims,
+                                                 BStrides,
+                                                 std::array<std::vector<ck::index_t>, 1>{DDims},
+                                                 std::array<std::vector<ck::index_t>, 1>{DStrides},
+                                                 EDims,
+                                                 EStrides,
+                                                 Pass{},
+                                                 Pass{},
+                                                 Bilinear{1.f, 1.f});
+        return contraction.IsSupportedArgument(argument);
+    }
+};
+
+template <typename DataTypeA,
+          typename DataTypeB,
+          typename DataTypeC,
+          typename DataTypeD,
+          ck::index_t NumDim>
+class ContractionDeviceOpWrapper
+{
+
+    protected:
+    using DeviceOp = ck::tensor_operation::device::DeviceContractionMultipleD<NumDim,
+                                                                              NumDim,
+                                                                              NumDim,
+                                                                              DataTypeA,
+                                                                              DataTypeB,
+                                                                              ck::Tuple<DataTypeC>,
+                                                                              DataTypeD,
+                                                                              Pass,
+                                                                              Pass,
+                                                                              Bilinear>;
+
+    public:
+    bool IsSupportedInstance(std::vector<ck::index_t>& Dims,
+                             std::vector<ck::index_t>& Strides) const
+    {
+
+        bool supported     = false;
+        const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+        for(auto& op_ptr : op_ptrs)
+        {
+            auto argument_ptr =
+                op_ptr->MakeArgumentPointer(nullptr,
+                                            nullptr,
+                                            std::array<const void*, 1>{nullptr},
+                                            nullptr,
+                                            Dims,
+                                            Strides,
+                                            Dims,
+                                            Strides,
+                                            std::array<std::vector<ck::index_t>, 1>{Dims},
+                                            std::array<std::vector<ck::index_t>, 1>{Strides},
+                                            Dims,
+                                            Strides,
+                                            Pass{},
+                                            Pass{},
+                                            Bilinear{1.f, 1.f});
+
+            supported = supported || op_ptr->IsSupportedArgument(argument_ptr.get());
+        }
+        return supported;
+    }
+};
+
+TEST(TestContractionInterface, IncorrectNumDims)
+{
+    std::vector<std::vector<ck::index_t>> Dims    = {{4, 4}, {4, 4, 4, 4}, {4, 4, 4, 4, 4, 4}};
+    std::vector<std::vector<ck::index_t>> Strides = {{1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1, 1, 1}};
+    ContractionDeviceOpWrapper<F32, F32, F32, F32, 1> wrapper_1d;
+    ContractionDeviceOpWrapper<F32, F32, F32, F32, 2> wrapper_2d;
+    ContractionDeviceOpWrapper<F32, F32, F32, F32, 3> wrapper_3d;
+    EXPECT_FALSE(wrapper_1d.IsSupportedInstance(Dims[0], Strides[0]));
+    EXPECT_TRUE(wrapper_2d.IsSupportedInstance(Dims[1], Strides[1]));
+    EXPECT_FALSE(wrapper_3d.IsSupportedInstance(Dims[2], Strides[2]));
+}
+
+TEST(TestContractionInterface, IncorrectDataTypes)
+{
+    std::vector<ck::index_t> Dims    = {4, 4, 4, 4};
+    std::vector<ck::index_t> Strides = {64, 16, 4, 1};
+    ContractionDeviceOpWrapper<F32, F32, F64, F64, 2> wrapper_1;
+    ContractionDeviceOpWrapper<F64, F64, F32, F32, 2> wrapper_2;
+    EXPECT_FALSE(wrapper_1.IsSupportedInstance(Dims, Strides));
+    EXPECT_FALSE(wrapper_2.IsSupportedInstance(Dims, Strides));
+}
+
+TEST(TestContractionSupportedArgs, ABMemoryAccess)
+{
+    std::vector<ck::index_t> Dims           = {4, 4, 4, 4};
+    std::vector<ck::index_t> Strides        = {64, 16, 4, 1};
+    std::vector<ck::index_t> StridesM1      = {4, 1, 64, 16};
+    std::vector<ck::index_t> StridesK1      = {64, 16, 4, 1};
+    std::vector<ck::index_t> InvalidStrides = {4, 4, 4, 4};
+    // Memory access to A
+    ContractionInstanceWrapper<1, 2, 4> wrapperA1;
+    ContractionInstanceWrapper<2, 2, 4> wrapperA2;
+    EXPECT_FALSE(
+        wrapperA1.isSupported(Dims, Dims, Dims, Dims, InvalidStrides, Strides, Strides, Strides));
+    EXPECT_FALSE(
+        wrapperA2.isSupported(Dims, Dims, Dims, Dims, InvalidStrides, Strides, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperA1.isSupported(Dims, Dims, Dims, Dims, StridesM1, Strides, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperA2.isSupported(Dims, Dims, Dims, Dims, StridesK1, Strides, Strides, Strides));
+    // Memory access to B
+    ContractionInstanceWrapper<2, 1, 4> wrapperB1;
+    ContractionInstanceWrapper<2, 2, 4> wrapperB2;
+    EXPECT_FALSE(
+        wrapperB1.isSupported(Dims, Dims, Dims, Dims, Strides, InvalidStrides, Strides, Strides));
+    EXPECT_FALSE(
+        wrapperB2.isSupported(Dims, Dims, Dims, Dims, Strides, InvalidStrides, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperB1.isSupported(Dims, Dims, Dims, Dims, Strides, StridesM1, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperB2.isSupported(Dims, Dims, Dims, Dims, Strides, StridesK1, Strides, Strides));
+}
+
+TEST(TestContractionSupportedArgs, DEMemoryAccess)
+{
+    std::vector<ck::index_t> Dims           = {4, 4, 4, 4};
+    std::vector<ck::index_t> Strides        = {64, 16, 4, 1};
+    std::vector<ck::index_t> InvalidStrides = {64, 16, 1, 4};
+    ContractionInstanceWrapper<2, 2, 4> wrapper;
+    // Memory access to D
+    EXPECT_FALSE(
+        wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, InvalidStrides, Strides));
+    EXPECT_TRUE(wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, Strides, Strides));
+    // Memory access to E
+    EXPECT_FALSE(
+        wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, Strides, InvalidStrides));
+    EXPECT_TRUE(wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, Strides, Strides));
+}
diff --git a/test/conv_util/conv_util.cpp b/test/conv_util/conv_util.cpp
index 73797a7169e78cfeabac101ac0c7da366187edbf..6922bbbcc73dca4d0bc9e98ae49e8b504adb860a 100644
--- a/test/conv_util/conv_util.cpp
+++ b/test/conv_util/conv_util.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <string>
diff --git a/test/convnd_bwd_data/CMakeLists.txt b/test/convnd_bwd_data/CMakeLists.txt
index 16ca4de8727878b25c75357e4ccff9bbc0517955..f734b46f53ef886e096b77d529f15644b3d7b96d 100644
--- a/test/convnd_bwd_data/CMakeLists.txt
+++ b/test/convnd_bwd_data/CMakeLists.txt
@@ -1,2 +1,9 @@
-add_gtest_executable(test_convnd_bwd_data convnd_bwd_data.cpp)
-target_link_libraries(test_convnd_bwd_data PRIVATE utility device_conv1d_bwd_data_instance device_conv2d_bwd_data_instance device_conv3d_bwd_data_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_gtest_executable(test_convnd_bwd_data convnd_bwd_data.cpp)
+   target_link_libraries(test_convnd_bwd_data PRIVATE utility device_conv1d_bwd_data_instance device_conv2d_bwd_data_instance device_conv3d_bwd_data_instance)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/convnd_bwd_data/convnd_bwd_data.cpp b/test/convnd_bwd_data/convnd_bwd_data.cpp
index 70231d42ae5f9eb9a368c643b2cf4041ae5fac4e..9d2b6cf5770144e25b21874e6f4da93c523e76bb 100644
--- a/test/convnd_bwd_data/convnd_bwd_data.cpp
+++ b/test/convnd_bwd_data/convnd_bwd_data.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/convnd_fwd/CMakeLists.txt b/test/convnd_fwd/CMakeLists.txt
index 97e170d85118b641a8338cde918bf60bda9a3803..745aceffc932389c8a8b2b35f384dcafa6f5691f 100644
--- a/test/convnd_fwd/CMakeLists.txt
+++ b/test/convnd_fwd/CMakeLists.txt
@@ -1,2 +1,9 @@
-add_gtest_executable(test_convnd_fwd convnd_fwd.cpp)
-target_link_libraries(test_convnd_fwd PRIVATE utility device_conv2d_fwd_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_gtest_executable(test_convnd_fwd convnd_fwd.cpp)
+   target_link_libraries(test_convnd_fwd PRIVATE utility device_conv2d_fwd_instance)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/convnd_fwd/convnd_fwd.cpp b/test/convnd_fwd/convnd_fwd.cpp
index a1921a9bfbe0f1221ddada95c75835ab9c942292..fe8798ceb8b1086d7bea43008194845a47f6db43 100644
--- a/test/convnd_fwd/convnd_fwd.cpp
+++ b/test/convnd_fwd/convnd_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/data_type/CMakeLists.txt b/test/data_type/CMakeLists.txt
index 088fbfec719f9e948686637fe9bec280f9926bd2..2b63727f19153dcc493bbb7ce4545b136a68b7db 100644
--- a/test/data_type/CMakeLists.txt
+++ b/test/data_type/CMakeLists.txt
@@ -2,3 +2,6 @@ if (USE_BITINT_EXTENSION_INT4)
   add_gtest_executable(test_int4 int4.cpp)
   target_link_libraries(test_int4 PRIVATE utility)
 endif()
+
+add_gtest_executable(test_fp8 fp8.cpp)
+target_link_libraries(test_fp8 PRIVATE utility)
diff --git a/test/data_type/fp8.cpp b/test/data_type/fp8.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..5004fe952775a74cc6b9736d1e34a8675d30187f
--- /dev/null
+++ b/test/data_type/fp8.cpp
@@ -0,0 +1,123 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/type_convert.hpp"
+
+using ck::f8_convert_sr;
+using ck::f8_t;
+using ck::half_t;
+using ck::type_convert;
+
+TEST(FP8, NumericLimits)
+{
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Min(), 0x08);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Max(), 0x77);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Lowest(), 0xF7);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::QuietNaN(), 0x80);
+}
+
+TEST(FP8, ConvertFP32Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to fp8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(type_convert<f8_t>(0.0f)), abs_tol);
+    // convert minimal float to fp8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal f8_t to float and check if equal to 240.0
+    ASSERT_NEAR(240.0f, type_convert<float>(type_convert<f8_t>(240.0f)), abs_tol);
+    // convert maximal float to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(240.0f,
+                type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to f8_t and check if it is qNan
+    ASSERT_NEAR(0x80, type_convert<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
+    // positive float value to fp8 and back, check if holds
+    float pos_float = 0.0078125f;
+    ASSERT_NEAR(pos_float, type_convert<float>(type_convert<f8_t>(pos_float)), abs_tol);
+    // negative float value to fp8 and back, check if holds
+    float neg_float = -0.0156250f;
+    ASSERT_NEAR(neg_float, type_convert<float>(type_convert<f8_t>(neg_float)), abs_tol);
+}
+
+TEST(FP8, ConvertFP32Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to fp8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(f8_convert_sr<f8_t>(0.0f)), abs_tol);
+    // convert minimal float to fp8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(f8_convert_sr<f8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal f8_t to float and check if equal to 240.0
+    ASSERT_NEAR(240.0f, type_convert<float>(f8_convert_sr<f8_t>(240.0f)), abs_tol);
+    // convert maximal float to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(240.0f,
+                type_convert<float>(f8_convert_sr<f8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to f8_t and check if it is qNan
+    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
+    // positive float value to fp8 and back, check if holds
+    float pos_float = 0.0078125f;
+    ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_sr<f8_t>(pos_float)), abs_tol);
+    // negative float value to fp8 and back, check if holds
+    float neg_float = -0.0156250f;
+    ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_sr<f8_t>(neg_float)), abs_tol);
+}
+
+TEST(FP8, ConvertFP16Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(type_convert<f8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal f8_t to fp16 and check if equal to 240.0
+    ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(type_convert<f8_t>(half_t{240.0})), abs_tol);
+    // convert maximal fp16 to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(half_t{240.0},
+                type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
+    ASSERT_NEAR(0x80, type_convert<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
+    // positive fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<f8_t>(pos_half)), abs_tol);
+    // negative fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.0156250};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<f8_t>(neg_half)), abs_tol);
+}
+
+TEST(FP8, ConvertFP16Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(f8_convert_sr<f8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal f8_t to fp16 and check if equal to 240.0
+    ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(f8_convert_sr<f8_t>(half_t{240.0})), abs_tol);
+    // convert maximal fp16 to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(half_t{240.0},
+                type_convert<half_t>(f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
+    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
+    // positive fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_sr<f8_t>(pos_half)), abs_tol);
+    // negative fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.0156250};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_sr<f8_t>(neg_half)), abs_tol);
+}
diff --git a/test/data_type/int4.cpp b/test/data_type/int4.cpp
index 252a450bf96896e677cbc5f402ccfe722192d9bc..07549c1c48bb807c9e50520e173d528bf27b3bc9 100644
--- a/test/data_type/int4.cpp
+++ b/test/data_type/int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <bitset>
 #include <cinttypes>
diff --git a/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp b/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp
index 403881b3cc442d0c16779b99ac678b7e30111cd4..d5ce77dc2b956b02439a42eaa341285217780ac8 100644
--- a/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp
+++ b/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_elementwise_layernorm_impl.hpp"
@@ -23,7 +23,7 @@ class TestElementwiseLayernorm : public ::testing::Test
     {
         // M, N
         std::vector<std::vector<ck::index_t>> lengths = {
-            {1, 1}, {25, 16}, {39, 777}, {100, 200}, {1024, 1024}, {48 * 256, 2048}};
+            {1, 1}, {25, 16}, {39, 777}, {100, 200}, {1024, 1024}, {48 * 256, 2048}, {4096, 8192}};
 
         for(auto length : lengths)
         {
diff --git a/test/gemm/CMakeLists.txt b/test/gemm/CMakeLists.txt
index c427586bb7933c2affed0c54b043f4e985346928..186b06d9117072dd7ecb7f74fc58461795b412c5 100644
--- a/test/gemm/CMakeLists.txt
+++ b/test/gemm/CMakeLists.txt
@@ -18,6 +18,7 @@ add_library(gemm_standalone_xdl_fp16_instances STATIC
     instance/gemm_f16_nn_instance.cpp
     instance/gemm_f16_nt_instance.cpp
     instance/gemm_f16_tn_instance.cpp
+    instance/gemm_wavelet_f16_tn_instance.cpp
     instance/gemm_f16_tt_instance.cpp
 )
 add_test_executable(test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp)
diff --git a/test/gemm/gemm_bf16.cpp b/test/gemm/gemm_bf16.cpp
index 5290d466323277daa4e61af1a8273b816a790120..cde5c45aeab59453488ea89e3942550531b65b87 100644
--- a/test/gemm/gemm_bf16.cpp
+++ b/test/gemm/gemm_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_fp16.cpp b/test/gemm/gemm_fp16.cpp
index 92e225def29623c7dcfc8590a5f805052dedb54d..cad250c6fb4d9eb935653d79999db3ff77a6932e 100644
--- a/test/gemm/gemm_fp16.cpp
+++ b/test/gemm/gemm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_fp32.cpp b/test/gemm/gemm_fp32.cpp
index 5d8c4881b621504fe621d056cbcf68bcb04622fc..c35aa77ea7dad4a7bf56e616d3456be932f44d8e 100644
--- a/test/gemm/gemm_fp32.cpp
+++ b/test/gemm/gemm_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_fp64.cpp b/test/gemm/gemm_fp64.cpp
index 85d7f95bf4ad9ea8df19fdf2600c191f7eabcbf2..e67c8ba4f32331619fa5979c0e66226253993af1 100644
--- a/test/gemm/gemm_fp64.cpp
+++ b/test/gemm/gemm_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_int8.cpp b/test/gemm/gemm_int8.cpp
index e73b22ce9c847f704fe55ad31cc83661b1ebab87..6ece05e306e66cb1013a846f3bdfda6ba4a195f4 100644
--- a/test/gemm/gemm_int8.cpp
+++ b/test/gemm/gemm_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_standalone_xdl_fp16.cpp b/test/gemm/gemm_standalone_xdl_fp16.cpp
index 8f5a5c557cdc4694895b67f18d6df293dece771e..201a49dcd390090c056ec911f70fee9bd61b61af 100644
--- a/test/gemm/gemm_standalone_xdl_fp16.cpp
+++ b/test/gemm/gemm_standalone_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_util.hpp"
 
@@ -10,6 +10,7 @@
 #include "gemm_f16_nt_instance.hpp"
 #include "gemm_f16_tn_instance.hpp"
 #include "gemm_f16_tt_instance.hpp"
+#include "gemm_wavelet_f16_tn_instance.hpp"
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -74,6 +75,10 @@ int main(int argc, char* argv[])
     {GemmParams{2048, 1664, 4096}, LayoutConfig{true, false, true}, add_gemm_f16_tn_256x128},
     {GemmParams{1024, 1664, 4096}, LayoutConfig{true, false, true}, add_gemm_f16_tn_128x128},
     {GemmParams{1024,  832, 4096}, LayoutConfig{true, false, true}, add_gemm_f16_tn_128x64},
+    {GemmParams{2048, 3328, 4096}, LayoutConfig{true, false, true}, add_gemm_wavelet_f16_tn_256x256},
+    {GemmParams{2048, 1664, 4096}, LayoutConfig{true, false, true}, add_gemm_wavelet_f16_tn_256x128},
+    {GemmParams{1024, 1664, 4096}, LayoutConfig{true, false, true}, add_gemm_wavelet_f16_tn_128x128},
+    {GemmParams{1024,  832, 4096}, LayoutConfig{true, false, true}, add_gemm_wavelet_f16_tn_128x64},
     {GemmParams{2048, 3328, 4096}, LayoutConfig{true, true, true}, add_gemm_f16_tt_256x256},
     {GemmParams{2048, 1664, 4096}, LayoutConfig{true, true, true}, add_gemm_f16_tt_256x128},
     {GemmParams{1024, 1664, 4096}, LayoutConfig{true, true, true}, add_gemm_f16_tt_128x128},
diff --git a/test/gemm/gemm_util.hpp b/test/gemm/gemm_util.hpp
index 9057c0af896891107683605575ee17aeb460c6cf..6c46f4ee89539b1e687b79f40803cfafa62528fc 100644
--- a/test/gemm/gemm_util.hpp
+++ b/test/gemm/gemm_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/test/gemm/instance/gemm_f16_nn_instance.cpp b/test/gemm/instance/gemm_f16_nn_instance.cpp
index 4d65c5876cd92a04cf04a99d3f3655e50c586a8d..9016257f131ffa4a130c009256e66b1d5abe709a 100644
--- a/test/gemm/instance/gemm_f16_nn_instance.cpp
+++ b/test/gemm/instance/gemm_f16_nn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_nn_instance.hpp b/test/gemm/instance/gemm_f16_nn_instance.hpp
index 5ae3928dc972f53eb1bb963cd2130e110ff19ae0..e174b99a1d8156540bdc3a0fcd16bb9870e96204 100644
--- a/test/gemm/instance/gemm_f16_nn_instance.hpp
+++ b/test/gemm/instance/gemm_f16_nn_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_nt_instance.cpp b/test/gemm/instance/gemm_f16_nt_instance.cpp
index 431ff1e62e79fc126deac89194996f2db3e2cb01..27103b88d4c25cd8658e8e58f2b0616377ccb99c 100644
--- a/test/gemm/instance/gemm_f16_nt_instance.cpp
+++ b/test/gemm/instance/gemm_f16_nt_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_nt_instance.hpp b/test/gemm/instance/gemm_f16_nt_instance.hpp
index 99f9ffba4562cd03baff678302e0900a1b842e32..c624425e69573e8293bb9b667f35cfee070c2806 100644
--- a/test/gemm/instance/gemm_f16_nt_instance.hpp
+++ b/test/gemm/instance/gemm_f16_nt_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tn_instance.cpp b/test/gemm/instance/gemm_f16_tn_instance.cpp
index 6f5dbc311ebb81501eedc218b43003c777424be2..5b11f4dad9000672b1c2cac01cdb5f1eb4a3ad34 100644
--- a/test/gemm/instance/gemm_f16_tn_instance.cpp
+++ b/test/gemm/instance/gemm_f16_tn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tn_instance.hpp b/test/gemm/instance/gemm_f16_tn_instance.hpp
index 62388aeb398dff03c1f997c5ece4d53560492abd..563e10600adde840ac967f5e980362d53e6460e9 100644
--- a/test/gemm/instance/gemm_f16_tn_instance.hpp
+++ b/test/gemm/instance/gemm_f16_tn_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tt_instance.cpp b/test/gemm/instance/gemm_f16_tt_instance.cpp
index b6ef5b1cd21a5314c83bedfcdef2d9c3d00906b9..9032150f0c5b7449244c23beea13df71d1e91ea5 100644
--- a/test/gemm/instance/gemm_f16_tt_instance.cpp
+++ b/test/gemm/instance/gemm_f16_tt_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tt_instance.hpp b/test/gemm/instance/gemm_f16_tt_instance.hpp
index 9d75b4e48cbba7905c3eebe2b30e8c663e31cbeb..62914d7ac22cb9a7ece1c321ef1a72368788ee2f 100644
--- a/test/gemm/instance/gemm_f16_tt_instance.hpp
+++ b/test/gemm/instance/gemm_f16_tt_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp b/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..23ff89da6005b9c612669c3932a763ef65433f19
--- /dev/null
+++ b/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
@@ -0,0 +1,96 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "gemm_wavelet_f16_tn_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+using gemm_f16_tn_256x256 = std::tuple<
+    // clang-format off
+    //#####################             | ALayout| BLayout| CLayout| AData| BData| AccData| CShuffle| CData|           A|           B|           C|           GEMM| NumGemmK| ABBlockTransfer|       BlockGemm|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+    //#####################             |        |        |        |  Type|  Type|    Type| DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+    //#####################             |        |        |        |      |      |        |         |      |   Operation|   Operation|   Operation|               |    Stage|                |                |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+    //#####################             |        |        |        |      |      |        |         |      |            |            |            |               |         |                |                |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    DeviceGemm_Xdl_WaveletModel_CShuffle<     Row,     Col,     Row,   F16,   F16,     F32,     F16,    F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,             256,             256,   256,   256,    32,   8,   8,   32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+using gemm_f16_tn_256x128 = std::tuple<
+    // clang-format off
+    //#####################             | ALayout| BLayout| CLayout| AData| BData| AccData| CShuffle| CData|           A|           B|           C|           GEMM| NumGemmK| ABBlockTransfer|       BlockGemm|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+    //#####################             |        |        |        |  Type|  Type|    Type| DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+    //#####################             |        |        |        |      |      |        |         |      |   Operation|   Operation|   Operation|               |    Stage|                |                |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+    //#####################             |        |        |        |      |      |        |         |      |            |            |            |               |         |                |                |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    DeviceGemm_Xdl_WaveletModel_CShuffle<     Row,     Col,     Row,   F16,   F16,     F32,      F16,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,             256,             256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+using gemm_f16_tn_128x128 = std::tuple<
+    // clang-format off
+    //#####################             | ALayout| BLayout| CLayout| AData| BData| AccData| CShuffle| CData|           A|           B|           C|           GEMM| NumGemmK| ABBlockTransfer|       BlockGemm|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+    //#####################             |        |        |        |  Type|  Type|    Type| DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+    //#####################             |        |        |        |      |      |        |         |      |   Operation|   Operation|   Operation|               |    Stage|                |                |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+    //#####################             |        |        |        |      |      |        |         |      |            |            |            |               |         |                |                |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    DeviceGemm_Xdl_WaveletModel_CShuffle<     Row,     Col,     Row,   F16,   F16,     F32,      F16,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,             256,             256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+using gemm_f16_tn_128x64 = std::tuple<
+    // clang-format off
+    //#####################             | ALayout| BLayout| CLayout| AData| BData| AccData| CShuffle| CData|           A|           B|           C|           GEMM| NumGemmK| ABBlockTransfer|       BlockGemm|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+    //#####################             |        |        |        |  Type|  Type|    Type| DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+    //#####################             |        |        |        |      |      |        |         |      |   Operation|   Operation|   Operation|               |    Stage|                |                |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+    //#####################             |        |        |        |      |      |        |         |      |            |            |            |               |         |                |                |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    DeviceGemm_Xdl_WaveletModel_CShuffle<     Row,     Col,     Row,   F16,   F16,     F32,      F16,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,             256,             256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_gemm_wavelet_f16_tn_256x256(std::vector<std::unique_ptr<BaseOperator>>& instances)
+{
+    add_device_operation_instances(instances, gemm_f16_tn_256x256{});
+}
+
+void add_gemm_wavelet_f16_tn_256x128(std::vector<std::unique_ptr<BaseOperator>>& instances)
+{
+    add_device_operation_instances(instances, gemm_f16_tn_256x128{});
+}
+
+void add_gemm_wavelet_f16_tn_128x128(std::vector<std::unique_ptr<BaseOperator>>& instances)
+{
+    add_device_operation_instances(instances, gemm_f16_tn_128x128{});
+}
+
+void add_gemm_wavelet_f16_tn_128x64(std::vector<std::unique_ptr<BaseOperator>>& instances)
+{
+    add_device_operation_instances(instances, gemm_f16_tn_128x64{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp b/test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..ef269d78ee5e11c8813b2f4787583c34530d7b2d
--- /dev/null
+++ b/test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <memory>
+#include <vector>
+
+#include "include/ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_gemm_wavelet_f16_tn_256x256(std::vector<std::unique_ptr<BaseOperator>>& instances);
+
+void add_gemm_wavelet_f16_tn_256x128(std::vector<std::unique_ptr<BaseOperator>>& instances);
+
+void add_gemm_wavelet_f16_tn_128x128(std::vector<std::unique_ptr<BaseOperator>>& instances);
+
+void add_gemm_wavelet_f16_tn_128x64(std::vector<std::unique_ptr<BaseOperator>>& instances);
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/test/gemm/run_gemm_test.inc b/test/gemm/run_gemm_test.inc
index ec27729b3c9f02148cdd5537fe7f12e7421a7a4d..d208bb5a7b95a1775815dfa95f55befa5c401ea9 100644
--- a/test/gemm/run_gemm_test.inc
+++ b/test/gemm/run_gemm_test.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run_gemm_test()
 {
diff --git a/test/gemm_layernorm/CMakeLists.txt b/test/gemm_layernorm/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..56b8d7737adf6db589cf15c4b615775843718029
--- /dev/null
+++ b/test/gemm_layernorm/CMakeLists.txt
@@ -0,0 +1,11 @@
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(test_gemm_layernorm)
+   add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp)
+   target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
+   add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp b/test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..3f059968784cd908a2969221d52ff89a2db1400f
--- /dev/null
+++ b/test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp
@@ -0,0 +1,77 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_gemm_add_relu_add_layernorm_impl.hpp"
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using F16 = ck::half_t;
+using F32 = float;
+using ck::index_t;
+
+template <typename Tuple>
+class TestGemmAddReluAddLayernorm : public ::testing::Test
+{
+    protected:
+    using ADataType        = std::tuple_element_t<0, Tuple>;
+    using BDataType        = std::tuple_element_t<1, Tuple>;
+    using AccDataType      = std::tuple_element_t<2, Tuple>;
+    using D0DataType       = std::tuple_element_t<3, Tuple>;
+    using D1DataType       = std::tuple_element_t<4, Tuple>;
+    using EMeanVarDataType = std::tuple_element_t<5, Tuple>;
+    using GammaDataType    = std::tuple_element_t<6, Tuple>;
+    using BetaDataType     = std::tuple_element_t<7, Tuple>;
+    using HDataType        = std::tuple_element_t<8, Tuple>;
+    using ALayout          = std::tuple_element_t<9, Tuple>;
+    using BLayout          = std::tuple_element_t<10, Tuple>;
+    using D0Layout         = std::tuple_element_t<11, Tuple>;
+    using D1Layout         = std::tuple_element_t<12, Tuple>;
+    using HLayout          = std::tuple_element_t<13, Tuple>;
+
+    void Run()
+    {
+        std::vector<std::vector<ck::index_t>> lengths = {
+            {1024, 1024, 1024}, {2048, 640, 640}, {1, 1, 1}};
+
+        for(auto length : lengths)
+        {
+            int M        = length[0];
+            int N        = length[1];
+            int K        = length[2];
+            int StrideA  = ck::is_same_v<ALayout, Row> ? K : M;
+            int StrideB  = ck::is_same_v<BLayout, Row> ? N : K;
+            int StrideD0 = 0;
+            int StrideD1 = ck::is_same_v<D1Layout, Row> ? N : M;
+            int StrideH  = ck::is_same_v<HLayout, Row> ? N : M;
+
+            bool success = ck::profiler::profile_gemm_add_relu_add_layernorm_impl<ADataType,
+                                                                                  BDataType,
+                                                                                  AccDataType,
+                                                                                  D0DataType,
+                                                                                  D1DataType,
+                                                                                  EMeanVarDataType,
+                                                                                  GammaDataType,
+                                                                                  BetaDataType,
+                                                                                  HDataType,
+                                                                                  ALayout,
+                                                                                  BLayout,
+                                                                                  D0Layout,
+                                                                                  D1Layout,
+                                                                                  HLayout>(
+                true, 1, false, false, M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideH);
+
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes = ::testing::Types<
+    std::tuple<F16, F16, F32, F16, F16, F16, F16, F16, F16, Row, Row, Row, Row, Row>,
+    std::tuple<F16, F16, F32, F16, F16, F16, F16, F16, F16, Row, Col, Row, Row, Row>,
+    std::tuple<F16, F16, F32, F16, F16, F16, F16, F16, F16, Col, Row, Row, Row, Row>,
+    std::tuple<F16, F16, F32, F16, F16, F16, F16, F16, F16, Col, Col, Row, Row, Row>>;
+
+TYPED_TEST_SUITE(TestGemmAddReluAddLayernorm, KernelTypes);
+TYPED_TEST(TestGemmAddReluAddLayernorm, Test_FP16) { this->Run(); }
diff --git a/test/gemm_reduce/gemm_reduce_fp16.cpp b/test/gemm_reduce/gemm_reduce_fp16.cpp
index 029165ece125abaa99dc1605bfbc2467b02ecc26..35a149f52c9f82634b4f63f837cbb6bbc808870e 100644
--- a/test/gemm_reduce/gemm_reduce_fp16.cpp
+++ b/test/gemm_reduce/gemm_reduce_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/gemm_split_k/CMakeLists.txt b/test/gemm_split_k/CMakeLists.txt
index 793091e53c8755db82eefe3cbf0af40666f71300..caf30fca595122a4d79935ea3bdbc10d83f364d1 100644
--- a/test/gemm_split_k/CMakeLists.txt
+++ b/test/gemm_split_k/CMakeLists.txt
@@ -1,3 +1,9 @@
-add_test_executable(test_gemm_split_k gemm_split_k.cpp)
-target_link_libraries(test_gemm_split_k PRIVATE utility)
-target_link_libraries(test_gemm_split_k PRIVATE device_gemm_splitk_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_gtest_executable(test_gemm_splitk test_gemm_splitk.cpp)
+   target_link_libraries(test_gemm_splitk PRIVATE utility device_gemm_splitk_instance)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/gemm_split_k/gemm_split_k.cpp b/test/gemm_split_k/gemm_split_k.cpp
deleted file mode 100644
index 1edb5769c6919817740e376d7c018781bb3930e2..0000000000000000000000000000000000000000
--- a/test/gemm_split_k/gemm_split_k.cpp
+++ /dev/null
@@ -1,261 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <initializer_list>
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "ck/library/utility/host_gemm.hpp"
-
-enum struct GemmMatrixLayout
-{
-    MK_KN_MN, // 0
-    MK_NK_MN, // 1
-    KM_KN_MN, // 2
-    KM_NK_MN, // 3
-};
-
-template <typename T>
-static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
-{
-    float max_diff = 1e-6;
-
-    for(std::size_t i = 0; i < ref.mData.size(); ++i)
-    {
-        float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
-        if(max_diff < diff)
-        {
-            return false;
-        }
-    }
-
-    return true;
-}
-
-struct gemmArgs
-{
-    GemmMatrixLayout layout;
-    int M;
-    int N;
-    int K;
-    int StrideA;
-    int StrideB;
-    int StrideC;
-    int KBatch;
-};
-
-int test_gemm(const gemmArgs& args)
-{
-    using Row = ck::tensor_layout::gemm::RowMajor;
-    using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-    bool a_row_major, b_row_major, c_row_major;
-
-    switch(args.layout)
-    {
-    case GemmMatrixLayout::MK_KN_MN:
-        a_row_major = true;
-        b_row_major = true;
-        c_row_major = true;
-        break;
-    case GemmMatrixLayout::MK_NK_MN:
-        a_row_major = true;
-        b_row_major = false;
-        c_row_major = true;
-        break;
-    case GemmMatrixLayout::KM_KN_MN:
-        a_row_major = false;
-        b_row_major = true;
-        c_row_major = true;
-        break;
-    case GemmMatrixLayout::KM_NK_MN:
-        a_row_major = false;
-        b_row_major = false;
-        c_row_major = true;
-        break;
-    default: printf("not supported layout"); return 1;
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, bool row_major) {
-            using namespace ck::literals;
-
-            if(row_major)
-            {
-                return HostTensorDescriptor({row, col}, {stride, 1_uz});
-            }
-            else
-            {
-                return HostTensorDescriptor({row, col}, {1_uz, stride});
-            }
-        };
-
-    Tensor<float> a_m_k(f_host_tensor_descriptor(args.M, args.K, args.StrideA, a_row_major));
-    Tensor<float> b_k_n(f_host_tensor_descriptor(args.K, args.N, args.StrideB, b_row_major));
-    Tensor<float> c_m_n_host_result(
-        f_host_tensor_descriptor(args.M, args.N, args.StrideC, c_row_major));
-    Tensor<float> c_m_n_device_result(
-        f_host_tensor_descriptor(args.M, args.N, args.StrideC, c_row_major));
-
-    // init data
-    std::size_t num_thread = 1;
-    a_m_k.GenerateTensorValue(GeneratorTensor_2<float>{-5, 5}, num_thread);
-    b_k_n.GenerateTensorValue(GeneratorTensor_2<float>{-5, 5}, num_thread);
-    // set zero to c_device_buf
-    c_m_n_device_result.GenerateTensorValue(GeneratorTensor_0<float>{}, num_thread);
-
-    host_gemm_mk_kn_mn(a_m_k,
-                       b_k_n,
-                       c_m_n_host_result,
-                       ck::tensor_operation::element_wise::PassThrough{},
-                       ck::tensor_operation::element_wise::PassThrough{},
-                       ck::tensor_operation::element_wise::PassThrough{});
-
-    DeviceMem a_device_buf(sizeof(float) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(float) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_device_buf(sizeof(float) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_m_k.mData.data());
-    b_device_buf.ToDevice(b_k_n.mData.data());
-    c_device_buf.ToDevice(c_m_n_device_result.mData.data());
-
-    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
-        bool success = false;
-
-        using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK<decltype(a_layout),
-                                                                        decltype(b_layout),
-                                                                        decltype(c_layout),
-                                                                        float,
-                                                                        float,
-                                                                        float,
-                                                                        PassThrough,
-                                                                        PassThrough,
-                                                                        PassThrough>;
-
-        const auto gemm_ptrs =
-            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
-                DeviceOp>::GetInstances();
-
-        for(auto& gemm_ptr : gemm_ptrs)
-        {
-            auto argument_ptr =
-                gemm_ptr->MakeArgumentPointer(static_cast<float*>(a_device_buf.GetDeviceBuffer()),
-                                              static_cast<float*>(b_device_buf.GetDeviceBuffer()),
-                                              static_cast<float*>(c_device_buf.GetDeviceBuffer()),
-                                              args.M,
-                                              args.N,
-                                              args.K,
-                                              args.StrideA,
-                                              args.StrideB,
-                                              args.StrideC,
-                                              ck::tensor_operation::element_wise::PassThrough{},
-                                              ck::tensor_operation::element_wise::PassThrough{},
-                                              ck::tensor_operation::element_wise::PassThrough{},
-                                              args.KBatch);
-
-            auto invoker_ptr = gemm_ptr->MakeInvokerPointer();
-
-            if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
-            {
-                invoker_ptr->Run(argument_ptr.get());
-
-                c_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-                if(!check_out(c_m_n_host_result, c_m_n_device_result))
-                {
-                    success = false;
-                    break;
-                }
-                success = true;
-            }
-        }
-
-        return success;
-    };
-
-    bool success = false;
-
-    if(args.layout == GemmMatrixLayout::MK_KN_MN)
-    {
-        success = test(Row{}, Row{}, Row{});
-    }
-    else if(args.layout == GemmMatrixLayout::MK_NK_MN)
-    {
-        success = test(Row{}, Col{}, Row{});
-    }
-    else if(args.layout == GemmMatrixLayout::KM_KN_MN)
-    {
-        success = test(Col{}, Row{}, Row{});
-    }
-    else
-    {
-        success = test(Col{}, Col{}, Row{});
-    }
-
-    auto error_code = 0;
-    if(success)
-    {
-        std::cout << "test split k : Pass" << std::endl;
-    }
-    else
-    {
-        std::cout << "test split k: Fail " << std::endl;
-        error_code = -1; // test needs to report failure
-    }
-    return error_code;
-}
-
-int main(int argc, char* argv[])
-{
-    std::vector<gemmArgs> test_cases;
-    if(argc == 1)
-    {
-        test_cases = {{GemmMatrixLayout::MK_KN_MN, 1024, 1024, 1024, 1024, 1024, 1024, 2},
-                      {GemmMatrixLayout::MK_KN_MN, 1024, 1024, 1024, 1024, 1024, 1024, 8}};
-    }
-    else if(argc == 9)
-    {
-        const auto layout = static_cast<GemmMatrixLayout>(std::stoi(argv[1]));
-
-        const int M = std::stoi(argv[2]);
-        const int N = std::stoi(argv[3]);
-        const int K = std::stoi(argv[4]);
-
-        const int StrideA = std::stoi(argv[5]);
-        const int StrideB = std::stoi(argv[6]);
-        const int StrideC = std::stoi(argv[7]);
-        const int KBatch  = std::stoi(argv[8]);
-        test_cases        = {{layout, M, N, K, StrideA, StrideB, StrideC, KBatch}};
-    }
-    else
-    {
-        printf("arg1: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
-        printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
-        printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
-        printf("                     3: A[k, m] * B[n, k] = C[m, n])\n");
-        printf("arg2 to 7: M, N, K, StrideA, StrideB, StrideC KBatch\n");
-        return -1;
-    }
-    bool error = false;
-    for(const auto& kinder : test_cases)
-    {
-        error |= test_gemm(kinder);
-    }
-    return error ? 1 : 0;
-}
diff --git a/test/gemm_split_k/test_gemm_splitk.cpp b/test/gemm_split_k/test_gemm_splitk.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..9eba5bba37551f39ffe921406ba871cff8dab4aa
--- /dev/null
+++ b/test/gemm_split_k/test_gemm_splitk.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "gtest/gtest.h"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "test_gemm_splitk_util.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+namespace {
+
+template <typename X, typename Y>
+struct tuple_concat;
+
+template <typename... Xs, typename... Ys>
+struct tuple_concat<std::tuple<Xs...>, std::tuple<Ys...>>
+{
+    using type = std::tuple<Xs..., Ys...>;
+};
+
+} // namespace
+
+template <typename Tuple>
+class TestGemmSplitK_MK_KN
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Row, Row>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmSplitK_MK_NK
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Row, Col>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmSplitK_KM_KN
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Col, Row>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmSplitK_KM_NK
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Col, Col>, Tuple>::type>
+{
+};
+
+// clang-format off
+using KernelTypes = ::testing::Types<
+    //         ADataType, BDataType, CDataType
+    std::tuple<      F16,       F16,       F16>,
+    std::tuple<      F32,       F32,       F32>
+    >;
+// clang-format on
+
+TYPED_TEST_SUITE(TestGemmSplitK_MK_KN, KernelTypes);
+TYPED_TEST_SUITE(TestGemmSplitK_MK_NK, KernelTypes);
+TYPED_TEST_SUITE(TestGemmSplitK_KM_KN, KernelTypes);
+TYPED_TEST_SUITE(TestGemmSplitK_KM_NK, KernelTypes);
+
+#include "test_gemm_splitk_ut_cases.inc"
diff --git a/test/gemm_split_k/test_gemm_splitk_ut_cases.inc b/test/gemm_split_k/test_gemm_splitk_ut_cases.inc
new file mode 100644
index 0000000000000000000000000000000000000000..54b9c6c9e314870446b75b9946cd2980ab008457
--- /dev/null
+++ b/test/gemm_split_k/test_gemm_splitk_ut_cases.inc
@@ -0,0 +1,217 @@
+#pragma once
+
+TYPED_TEST(TestGemmSplitK_MK_KN, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_KN, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_KN, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_KN, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
diff --git a/test/gemm_split_k/test_gemm_splitk_util.hpp b/test/gemm_split_k/test_gemm_splitk_util.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..8243747a694141b373cd6d6896dd17213fd3192c
--- /dev/null
+++ b/test/gemm_split_k/test_gemm_splitk_util.hpp
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <string>
+#include <sstream>
+#include <tuple>
+#include <vector>
+#include <gtest/gtest.h>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "include/ck/utility/data_type.hpp"
+#include "profiler/profile_gemm_splitk_impl.hpp"
+
+namespace ck {
+namespace test {
+
+template <typename Tuple>
+class TestGemmSplitK : public testing::Test
+{
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using F32 = float;
+
+    protected:
+    using ALayout   = std::tuple_element_t<0, Tuple>;
+    using BLayout   = std::tuple_element_t<1, Tuple>;
+    using CLayout   = Row;
+    using ADataType = std::tuple_element_t<2, Tuple>;
+    using BDataType = std::tuple_element_t<3, Tuple>;
+    using CDataType = std::tuple_element_t<4, Tuple>;
+
+    public:
+    static constexpr bool verify_     = true;
+    static constexpr int init_method_ = 1; // decimal value initialization
+    static constexpr bool log_        = false;
+    static constexpr bool bench_      = false; // measure kernel performance
+    std::vector<int> k_batches_;
+
+    void SetUp() override { k_batches_ = {1, 2, 3, 5, 8}; }
+
+    void Run(const int M,
+             const int N,
+             const int K,
+             const int StrideA,
+             const int StrideB,
+             const int StrideC)
+    {
+        for(auto kb : k_batches_)
+        {
+            RunSingle(M, N, K, StrideA, StrideB, StrideC, kb);
+        }
+    }
+
+    void RunSingle(const int M,
+                   const int N,
+                   const int K,
+                   const int StrideA,
+                   const int StrideB,
+                   const int StrideC,
+                   int kbatch = 1)
+    {
+        bool pass = ck::profiler::profile_gemm_splitk_impl<ADataType,
+                                                           BDataType,
+                                                           F32,
+                                                           CDataType,
+                                                           ALayout,
+                                                           BLayout,
+                                                           CLayout>(
+            verify_, init_method_, log_, bench_, M, N, K, StrideA, StrideB, StrideC, kbatch);
+        EXPECT_TRUE(pass);
+    }
+};
+
+} // namespace test
+} // namespace ck
diff --git a/test/grouped_convnd_bwd_data/CMakeLists.txt b/test/grouped_convnd_bwd_data/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d7c9f465548f5e6756ed043bf28fb322dc71753d
--- /dev/null
+++ b/test/grouped_convnd_bwd_data/CMakeLists.txt
@@ -0,0 +1,6 @@
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
+    target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+    add_gtest_executable(test_grouped_convnd_bwd_data_interface test_grouped_convnd_bwd_data_interface.cpp)
+    target_link_libraries(test_grouped_convnd_bwd_data_interface PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+endif()
\ No newline at end of file
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..94808669a321d1ebd766dd62f05d18ba8ccbf47b
--- /dev/null
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
+
+template <typename Tuple>
+class TestGroupedConvndBwdData : public ::testing::Test
+{
+    protected:
+    using DataType  = std::tuple_element_t<0, Tuple>;
+    using OutLayout = std::tuple_element_t<1, Tuple>;
+    using WeiLayout = std::tuple_element_t<2, Tuple>;
+    using InLayout  = std::tuple_element_t<3, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+
+    template <ck::index_t NDimSpatial>
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_grouped_conv_bwd_data_impl<NDimSpatial,
+                                                                            OutLayout,
+                                                                            WeiLayout,
+                                                                            InLayout,
+                                                                            DataType,
+                                                                            DataType,
+                                                                            DataType>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes = ::testing::Types<std::tuple<float, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<ck::half_t, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<ck::bhalf_t, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<float, NHWGK, GKYXC, NHWGC>,
+                                     std::tuple<ck::half_t, NHWGK, GKYXC, NHWGC>,
+                                     std::tuple<ck::bhalf_t, NHWGK, GKYXC, NHWGC>>;
+TYPED_TEST_SUITE(TestGroupedConvndBwdData, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdData, Test2D)
+{
+    this->conv_params.clear();
+
+    this->conv_params.push_back(
+        {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->template Run<2>();
+}
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..bc592ba665604f93bdcd3b545152376d14d55c88
--- /dev/null
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp
@@ -0,0 +1,178 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using DataType    = ck::half_t;
+using AccDataType = float;
+using Pass        = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+using ConvBackwardDataSpecialization =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization;
+
+static constexpr auto ConvBwdDataDefault   = ConvBackwardDataSpecialization::Default;
+static constexpr auto Filter1x1Stride1Pad0 = ConvBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+template <typename Tuple, ConvBackwardDataSpecialization ConvSpec>
+class TestGroupedConvndBwdData : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 2;
+
+    using OutLayout = std::tuple_element_t<0, Tuple>;
+    using WeiLayout = std::tuple_element_t<1, Tuple>;
+    using InLayout  = std::tuple_element_t<2, Tuple>;
+
+    // clang-format off
+    using GroupedConvBwdDataDeviceInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
+    // ######| NDimSpatial|   ALayout|   BLayout|   DsLayout|  ELayout|       AData|       BData|     AccData|         CShuffle|       DsData|      EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+    // ######|            |          |          |           |         |        Type|        Type|        Type|         DataType|         Type|       Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+    // ######|            |          |          |           |         |            |            |            |                 |             |           |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+    // ######|            |          |          |           |         |            |            |            |                 |             |           |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+    < NDimSpatial, OutLayout, WeiLayout, ck::Tuple<>, InLayout, DataType, DataType, AccDataType, DataType,   ck::Tuple<>, DataType, Pass, Pass,    Pass,  ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+
+    template <ck::index_t NDimSpatial>
+    bool Run()
+    {
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> out_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> in_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+        copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
+        copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
+        copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
+        copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
+        copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
+        copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
+        copy(conv_param.conv_filter_strides_, conv_filter_strides);
+        copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+        copy(conv_param.input_left_pads_, input_left_pads);
+        copy(conv_param.input_right_pads_, input_right_pads);
+
+        auto conv = GroupedConvBwdDataDeviceInstance{};
+
+        auto argument = conv.MakeArgument(nullptr,
+                                          nullptr,
+                                          std::array<const void*, 0>{},
+                                          nullptr,
+                                          out_lengths,
+                                          out_strides,
+                                          wei_lengths,
+                                          wei_strides,
+                                          {},
+                                          {},
+                                          in_lengths,
+                                          in_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          Pass{},
+                                          Pass{},
+                                          Pass{});
+        return conv.IsSupportedArgument(argument);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes =
+    ::testing::Types<std::tuple<GNHWK, GKYXC, GNHWC>, std::tuple<NHWGK, GKYXC, NHWGC>>;
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataDefault : public TestGroupedConvndBwdData<Tuple, ConvBwdDataDefault>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataFilter1x1
+    : public TestGroupedConvndBwdData<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataDefault, KernelTypes);
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataFilter1x1, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdDataFilter1x1, SpecializationCheck)
+{
+    // Check filter 3,3 instead of 1,1
+    this->conv_param  = {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check strides 2,2 instead of 1,1
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check with pad
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Supported version
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_TRUE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdDataDefault, VectorLoadCheck)
+{
+    // vector load for A
+    this->conv_param  = {2, 2, 128, 129, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+    // vector load for B, E, Ds
+    this->conv_param = {2, 2, 128, 128, 257, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+}
diff --git a/test/grouped_convnd_bwd_weight/CMakeLists.txt b/test/grouped_convnd_bwd_weight/CMakeLists.txt
index e2f0790c8b6b63381454ebd9db368e985714e659..f1a3f04b77a8ef4aceac5a6dc233419c8cc26c0c 100644
--- a/test/grouped_convnd_bwd_weight/CMakeLists.txt
+++ b/test/grouped_convnd_bwd_weight/CMakeLists.txt
@@ -1,2 +1,11 @@
-add_gtest_executable(test_grouped_convnd_bwd_weight grouped_convnd_bwd_weight.cpp) 
-target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_gtest_executable(test_grouped_convnd_bwd_weight test_grouped_convnd_bwd_weight.cpp)
+   target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+   add_gtest_executable(test_grouped_convnd_bwd_weight_interface test_grouped_convnd_bwd_weight_interface.cpp)
+   target_link_libraries(test_grouped_convnd_bwd_weight_interface PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
deleted file mode 100644
index 6545b6e5662bb971a33e2b0d158a3440636b35fe..0000000000000000000000000000000000000000
--- a/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
+++ /dev/null
@@ -1,91 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-#include <iostream>
-#include <initializer_list>
-#include <tuple>
-#include <vector>
-
-#include <gtest/gtest.h>
-
-#include "profiler/profile_grouped_conv_bwd_weight_impl.hpp"
-
-template <typename Tuple>
-class TestGroupedConvndBwdWeight : public ::testing::Test
-{
-    protected:
-    using DataType = std::tuple_element_t<0, Tuple>;
-    std::vector<ck::utils::conv::ConvParam> conv_params;
-    ck::index_t split_k{2};
-
-    template <ck::index_t NDimSpatial>
-    void Run()
-    {
-        for(auto& param : conv_params)
-        {
-            bool pass;
-            EXPECT_FALSE(conv_params.empty());
-            pass = ck::profiler::profile_grouped_conv_bwd_weight_impl<
-                NDimSpatial,
-                ck::tuple_element_t<NDimSpatial - 1,
-                                    ck::Tuple<ck::tensor_layout::convolution::GNWC,
-                                              ck::tensor_layout::convolution::GNHWC,
-                                              ck::tensor_layout::convolution::GNDHWC>>,
-                ck::tuple_element_t<NDimSpatial - 1,
-                                    ck::Tuple<ck::tensor_layout::convolution::GKXC,
-                                              ck::tensor_layout::convolution::GKYXC,
-                                              ck::tensor_layout::convolution::GKZYXC>>,
-                ck::tuple_element_t<NDimSpatial - 1,
-                                    ck::Tuple<ck::tensor_layout::convolution::GNWK,
-                                              ck::tensor_layout::convolution::GNHWK,
-                                              ck::tensor_layout::convolution::GNDHWK>>,
-                DataType,
-                DataType,
-                DataType>(true,  // do_verification
-                          1,     // init_method integer value
-                          false, // do_log
-                          false, // time_kernel
-                          param,
-                          split_k);
-            EXPECT_TRUE(pass);
-        }
-    }
-};
-
-using KernelTypes =
-    ::testing::Types<std::tuple<float>, std::tuple<ck::half_t>, std::tuple<ck::bhalf_t>>;
-TYPED_TEST_SUITE(TestGroupedConvndBwdWeight, KernelTypes);
-
-TYPED_TEST(TestGroupedConvndBwdWeight, Test1D)
-{
-    this->conv_params.clear();
-    this->conv_params.push_back({1, 4, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
-    this->conv_params.push_back({1, 4, 64, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
-    this->conv_params.push_back({1, 4, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
-    this->template Run<1>();
-}
-
-TYPED_TEST(TestGroupedConvndBwdWeight, Test2D)
-{
-    this->conv_params.clear();
-    this->conv_params.push_back(
-        {2, 4, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
-    this->conv_params.push_back(
-        {2, 4, 8, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    this->conv_params.push_back(
-        {2, 4, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
-    this->template Run<2>();
-}
-
-TYPED_TEST(TestGroupedConvndBwdWeight, Test3D)
-{
-    this->conv_params.clear();
-    this->conv_params.push_back(
-        {3, 4, 128, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    this->conv_params.push_back(
-        {3, 4, 8, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
-    this->conv_params.push_back(
-        {3, 4, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    this->template Run<3>();
-}
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..b6c79fa80b957d444e196272b230bd1e6f6e2211
--- /dev/null
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
@@ -0,0 +1,125 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+#include "profiler/profile_grouped_conv_bwd_weight_impl.hpp"
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight : public ::testing::Test
+{
+    protected:
+    using InDataType  = std::tuple_element_t<0, Tuple>;
+    using WeiDataType = std::tuple_element_t<1, Tuple>;
+    using OutDataType = std::tuple_element_t<2, Tuple>;
+    using InLayout    = std::tuple_element_t<3, Tuple>;
+    using WeiLayout   = std::tuple_element_t<4, Tuple>;
+    using OutLayout   = std::tuple_element_t<5, Tuple>;
+    using NDimSpatial = std::tuple_element_t<6, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+    ck::index_t split_k{2};
+
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_grouped_conv_bwd_weight_impl<NDimSpatial{},
+                                                                              InLayout,
+                                                                              WeiLayout,
+                                                                              OutLayout,
+                                                                              InDataType,
+                                                                              WeiDataType,
+                                                                              OutDataType>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param,
+                               split_k);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight1d : public TestGroupedConvndBwdWeight<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight2d : public TestGroupedConvndBwdWeight<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight3d : public TestGroupedConvndBwdWeight<Tuple>
+{
+};
+
+using namespace ck::tensor_layout::convolution;
+
+using KernelTypes1d = ::testing::Types<
+    std::tuple<float, float, float, GNWC, GKXC, GNWK, ck::Number<1>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, GNWC, GKXC, GNWK, ck::Number<1>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNWC, GKXC, GNWK, ck::Number<1>>>;
+using KernelTypes2d = ::testing::Types<
+    std::tuple<float, float, float, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
+    std::tuple<float, float, float, NHWGC, GKYXC, NHWGK, ck::Number<2>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, NHWGC, GKYXC, NHWGK, ck::Number<2>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, NHWGC, GKYXC, NHWGK, ck::Number<2>>>;
+using KernelTypes3d = ::testing::Types<
+    std::tuple<float, float, float, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>>;
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeight1d, KernelTypes1d);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeight2d, KernelTypes2d);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeight3d, KernelTypes3d);
+
+TYPED_TEST(TestGroupedConvndBwdWeight1d, Test1D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 2, 32, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
+    this->Run();
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeight2d, Test2D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {2, 2, 64, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back(
+        {2, 2, 4, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->Run();
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeight3d, Test3D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {3, 2, 16, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 2, 2, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->Run();
+}
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..57d4e4186c9f4a37ac51a9bba7566e5601abe820
--- /dev/null
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
@@ -0,0 +1,180 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using F16         = ck::half_t;
+using F32         = float;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+using ConvolutionBackwardWeightSpecialization =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization;
+
+static constexpr auto ConvBwdWeightDefault = ConvolutionBackwardWeightSpecialization::Default;
+static constexpr auto Filter1x1Stride1Pad0 =
+    ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <typename Tuple, ConvolutionBackwardWeightSpecialization ConvSpec>
+class TestGroupedConvndBwdWeight : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 2;
+
+    using InLayout  = std::tuple_element_t<2, Tuple>;
+    using WeiLayout = std::tuple_element_t<1, Tuple>;
+    using OutLayout = std::tuple_element_t<0, Tuple>;
+
+    // clang-format off
+    using GroupedConvBwdWeightDeviceInstance = ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle
+        //##########|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //##########|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //##########| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //##########|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        <         NDimSpatial,  InLayout, WeiLayout,OutLayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+    ck::index_t split_k{2};
+
+    template <ck::index_t NDimSpatial>
+    bool Run()
+    {
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> output_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); };
+
+        range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
+        range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
+        range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+        range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+        range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
+        range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
+        range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
+        range_copy(conv_param.input_left_pads_, begin(input_left_pads));
+        range_copy(conv_param.input_right_pads_, begin(input_right_pads));
+
+        auto conv = GroupedConvBwdWeightDeviceInstance{};
+
+        auto argument = conv.MakeArgument(nullptr,
+                                          nullptr,
+                                          nullptr,
+                                          conv_param.G_,
+                                          conv_param.N_,
+                                          conv_param.K_,
+                                          conv_param.C_,
+                                          input_spatial_lengths,
+                                          filter_spatial_lengths,
+                                          output_spatial_lengths,
+                                          input_strides,
+                                          output_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          PassThrough{},
+                                          PassThrough{},
+                                          PassThrough{},
+                                          split_k);
+        return conv.IsSupportedArgument(argument);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes =
+    ::testing::Types<std::tuple<GNHWK, GKYXC, GNHWC>, std::tuple<NHWGK, GKYXC, NHWGC>>;
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeightDefault
+    : public TestGroupedConvndBwdWeight<Tuple, ConvBwdWeightDefault>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeightFilter1x1
+    : public TestGroupedConvndBwdWeight<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightDefault, KernelTypes);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightFilter1x1, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdWeightFilter1x1, SpecializationCheck)
+{
+    // Check filter 3,3 instead of 1,1
+    this->conv_param  = {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check strides 2,2 instead of 1,1
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check with pad
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Supported version
+    this->conv_param = {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_TRUE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeightDefault, VectorLoadCheck)
+{
+    // vector load for A
+    this->conv_param  = {2, 2, 128, 129, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+    // vector load for B, E, Ds
+    this->conv_param = {2, 2, 128, 128, 257, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+}
diff --git a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp b/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
index 6df7f9969cb5d802ae28d0a7fdf03221c6c91c4f..4a804ef7f1af5abe4aa040867335043a726189d1 100644
--- a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
+++ b/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/grouped_gemm/CMakeLists.txt b/test/grouped_gemm/CMakeLists.txt
index 31a78733d38c14f6f328d1bdea3ea3cbd3341507..8c57b667e27287f51c79dc79b879506f5d4d0ba4 100644
--- a/test/grouped_gemm/CMakeLists.txt
+++ b/test/grouped_gemm/CMakeLists.txt
@@ -1,3 +1,14 @@
-add_test_executable(test_grouped_gemm_fp16 grouped_gemm_fp16.cpp)
-target_link_libraries(test_grouped_gemm_fp16 PRIVATE utility)
-target_link_libraries(test_grouped_gemm_fp16 PRIVATE device_grouped_gemm_instance)
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(test_grouped_gemm)
+   add_gtest_executable(test_grouped_gemm_splitk test_grouped_gemm_splitk.cpp)
+   add_gtest_executable(test_grouped_gemm_interface test_grouped_gemm_interface.cpp)
+   target_link_libraries(test_grouped_gemm_splitk PRIVATE utility device_grouped_gemm_instance)
+   target_link_libraries(test_grouped_gemm_interface PRIVATE utility device_grouped_gemm_instance)
+   
+   add_dependencies(test_grouped_gemm test_grouped_gemm_splitk test_grouped_gemm_interface)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/grouped_gemm/grouped_gemm_fp16.cpp b/test/grouped_gemm/grouped_gemm_fp16.cpp
deleted file mode 100644
index b3f7cca418f969a70d1e85fe569f90e2d91c4240..0000000000000000000000000000000000000000
--- a/test/grouped_gemm/grouped_gemm_fp16.cpp
+++ /dev/null
@@ -1,65 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-
-#include "profiler/profile_grouped_gemm_impl.hpp"
-
-namespace {
-
-using ADataType   = ck::half_t;
-using BDataType   = ck::half_t;
-using CDataType   = ck::half_t;
-using AccDataType = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <typename ALayout, typename BLayout, typename CLayout>
-bool TestGroupedGemm()
-{
-    int group_count = rand() % 10 + 1;
-
-    // GEMM shape
-    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
-    std::vector<const void*> p_a, p_b;
-    std::vector<void*> p_c;
-
-    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideCs;
-
-    for(int i = 0; i < group_count; i++)
-    {
-        Ms.push_back(256 + 256 * (rand() % 10));
-        Ns.push_back(256 + 256 * (rand() % 10));
-        Ks.push_back(128 + 128 * (rand() % 10));
-
-        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
-        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
-        StrideCs.push_back(std::is_same<Row, CLayout>::value ? Ns[i] : Ms[i]);
-    }
-
-    return ck::profiler::profile_grouped_gemm_impl<ADataType,
-                                                   BDataType,
-                                                   CDataType,
-                                                   AccDataType,
-                                                   ALayout,
-                                                   BLayout,
-                                                   CLayout>(
-        true, 1, false, 1, Ms, Ns, Ks, StrideAs, StrideBs, StrideCs);
-}
-
-} // anonymous namespace
-
-int main()
-{
-    bool res = true;
-
-    res = res && TestGroupedGemm<Row, Row, Row>();
-    res = res && TestGroupedGemm<Row, Col, Row>();
-    res = res && TestGroupedGemm<Col, Row, Row>();
-    res = res && TestGroupedGemm<Col, Col, Row>();
-
-    std::cout << "TestGroupedGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-
-    return res ? 0 : 1;
-}
diff --git a/test/grouped_gemm/test_grouped_gemm_interface.cpp b/test/grouped_gemm/test_grouped_gemm_interface.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ffa8840fc7d8ba6b8a1c1a39c62edc6edae58064
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_interface.cpp
@@ -0,0 +1,202 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <stdexcept>
+#include <vector>
+#include "gtest/gtest.h"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "test_grouped_gemm_util.hpp"
+
+class TestGGemmSplitKInterface_MKNKMN : public ::testing::Test
+{
+    protected:
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using ALayout = Row;
+    using BLayout = Col;
+    using ELayout = Row;
+
+    static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+    template <ck::tensor_operation::device::GemmSpecialization GemmSpec,
+              ck::index_t KPerBlock,
+              ck::index_t K1,
+              ck::index_t ABlockTransferSrcScalarPerVector,
+              ck::index_t BBlockTransferSrcScalarPerVector,
+              ck::index_t CDEBlockTransferScalarPerVector_NPerBlock>
+    using GGemmInstance =
+        ck::test::DeviceGroupedGemmSplitkInstanceWrapper<ALayout,
+                                                         BLayout,
+                                                         ELayout,
+                                                         GemmSpec,
+                                                         KPerBlock,
+                                                         K1,
+                                                         ABlockTransferSrcScalarPerVector,
+                                                         BBlockTransferSrcScalarPerVector,
+                                                         CDEBlockTransferScalarPerVector_NPerBlock>;
+
+    using DefaultGGemmInstance = GGemmInstance<GemmDefault, 32, 8, 4, 8, 8>;
+};
+
+TEST_F(TestGGemmSplitKInterface_MKNKMN, TileSize)
+{
+    std::vector<int> Ms{128, 256, 188, 512};
+    constexpr int N = 256;
+    constexpr int K = 128;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // M % MPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ms = std::vector<int>{256, 128, 128, 512};
+    Ns = std::vector<int>{256, 177, 128, 512};
+    // N % NPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
+
+TEST_F(TestGGemmSplitKInterface_MKNKMN, VectorLoadWidth)
+{
+    static constexpr auto GemmMNKPadding =
+        ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+    using PaddedGGemmInstance = GGemmInstance<GemmMNKPadding, 32, 8, 4, 8, 8>;
+
+    std::vector<int> Ms{128, 256, 256, 512};
+    constexpr int N = 256;
+    constexpr int K = 512;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // K % ABlockTransferSrcScalarPerVector
+    Ks = std::vector<int>{256, 177, 128, 512};
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ks = std::vector<int>{256, 164, 128, 512};
+    // K % BBlockTransferSrcScalarPerVector
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ks = std::vector<int>(4, 128);
+    Ns = std::vector<int>{256, 127, 128, 512};
+    // N % CBlockTransferScalarPerVector_NWaveNPerXDL
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
+
+TEST_F(TestGGemmSplitKInterface_MKNKMN, KLoops)
+{
+    std::vector<int> Ms{128, 256, 256, 512};
+    constexpr int N      = 256;
+    constexpr int K      = 128;
+    constexpr int kbatch = 4;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // kloops % 2
+    Ks = std::vector<int>{256, 512, 320, 768};
+    EXPECT_FALSE(
+        DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch));
+
+    // Not all gemms have same value for main_k0_block_loop!
+    Ks = std::vector<int>{256, 512, 512, 512};
+    EXPECT_THROW(DefaultGGemmInstance{}.Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch),
+                 std::runtime_error);
+}
+
+class TestGGemmSplitKInterface_KMKNNM : public ::testing::Test
+{
+    protected:
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using ALayout = Col;
+    using BLayout = Row;
+    using ELayout = Col;
+
+    static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+    template <ck::tensor_operation::device::GemmSpecialization GemmSpec,
+              ck::index_t KPerBlock,
+              ck::index_t K1,
+              ck::index_t ABlockTransferSrcScalarPerVector,
+              ck::index_t BBlockTransferSrcScalarPerVector,
+              ck::index_t CDEBlockTransferScalarPerVector_NPerBlock>
+    using GGemmInstance =
+        ck::test::DeviceGroupedGemmSplitkInstanceWrapper<ALayout,
+                                                         BLayout,
+                                                         ELayout,
+                                                         GemmSpec,
+                                                         KPerBlock,
+                                                         K1,
+                                                         ABlockTransferSrcScalarPerVector,
+                                                         BBlockTransferSrcScalarPerVector,
+                                                         CDEBlockTransferScalarPerVector_NPerBlock>;
+
+    using DefaultGGemmInstance = GGemmInstance<GemmDefault, 32, 8, 4, 8, 4>;
+};
+
+TEST_F(TestGGemmSplitKInterface_KMKNNM, TileSize)
+{
+    std::vector<int> Ms{128, 256, 188, 512};
+    constexpr int N = 256;
+    constexpr int K = 128;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // M % MPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ms = std::vector<int>{128, 256, 256, 512};
+    Ns = std::vector<int>{256, 177, 128, 512};
+    // N % NPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
+
+TEST_F(TestGGemmSplitKInterface_KMKNNM, VectorLoadWidth)
+{
+    static constexpr auto GemmMNKPadding =
+        ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+    using PaddedGGemmInstance = GGemmInstance<GemmMNKPadding, 32, 8, 2, 8, 4>;
+
+    std::vector<int> Ms{128, 256, 256, 512};
+    constexpr int N = 256;
+    constexpr int K = 512;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // M % ABlockTransferSrcScalarPerVector
+    Ms = std::vector<int>{256, 177, 128, 512};
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ms = std::vector<int>{128, 256, 256, 512};
+    Ns = std::vector<int>{256, 164, 128, 512};
+    // N % BBlockTransferSrcScalarPerVector
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ns = std::vector<int>{128, 256, 256, 512};
+    Ms = std::vector<int>{256, 130, 128, 512};
+    // M % CBlockTransferScalarPerVector_NWaveNPerXDL
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
diff --git a/test/grouped_gemm/test_grouped_gemm_splitk.cpp b/test/grouped_gemm/test_grouped_gemm_splitk.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..d9282fa924943c52d317f1fe89b45c90e54c3afb
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_splitk.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "gtest/gtest.h"
+#include "test_grouped_gemm_util.hpp"
+
+using F16 = ck::half_t;
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using RRR_F16_F16_F16 = ck::test::TestGroupedGemm<std::tuple<Row, Row, Row, F16, F16, F16>>;
+using RCR_F16_F16_F16 = ck::test::TestGroupedGemm<std::tuple<Row, Col, Row, F16, F16, F16>>;
+
+using RRR_F16_F16_F16_LargeK = ck::test::TestGroupedGemm<std::tuple<Row, Row, Row, F16, F16, F16>>;
+using RCR_F16_F16_F16_LargeK = ck::test::TestGroupedGemm<std::tuple<Row, Col, Row, F16, F16, F16>>;
+
+const std::vector<int> KBATCH{1, 2, 3, 5, 8};
+
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_MK_KN, RRR_F16_F16_F16, testing::ValuesIn(KBATCH));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_MK_NK, RCR_F16_F16_F16, testing::ValuesIn(KBATCH));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_LargeK_MK_KN,
+                         RRR_F16_F16_F16_LargeK,
+                         testing::Values(32, 64));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_LargeK_MK_NK,
+                         RCR_F16_F16_F16_LargeK,
+                         testing::Values(32, 64));
+
+#include "test_grouped_gemm_ut_cases.inc"
diff --git a/test/grouped_gemm/test_grouped_gemm_ut_cases.inc b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
new file mode 100644
index 0000000000000000000000000000000000000000..d94d140d97ecfb1ce0d879af747fc2badaba6ecc
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
@@ -0,0 +1,180 @@
+#pragma once
+
+TEST_P(RRR_F16_F16_F16, TinyCases)
+{
+    const std::vector<int> Ms{0, 1};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, SmallCases)
+{
+    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, MidCases)
+{
+    const std::vector<int> Ms{167, 183, 177, 153, 139, 204};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, Regular)
+{
+    const std::vector<int> Ms{64, 128, 256};
+    constexpr int N = 768;
+    constexpr int K = 320;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, MNKPadded)
+{
+    const std::vector<int> Ms{127, 150, 188, 210};
+    constexpr int N = 136;
+    constexpr int K = 280;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, TinyCases)
+{
+    const std::vector<int> Ms{0, 1};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, SmallCases)
+{
+    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, MidCases)
+{
+    const std::vector<int> Ms{167, 183, 177, 153, 139, 204};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, Regular)
+{
+    const std::vector<int> Ms{32, 64, 128, 256};
+    constexpr int N = 768;
+    constexpr int K = 320;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, MNKPadded)
+{
+    const std::vector<int> Ms{127, 150, 188, 210};
+    constexpr int N = 136;
+    constexpr int K = 280;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16_LargeK, TestLargeKBatch)
+{
+    const std::vector<int> Ms{188, 210};
+    constexpr int N = 768;
+    constexpr int K = 4096;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16_LargeK, TestLargeKBatch)
+{
+    const std::vector<int> Ms{188, 210};
+    constexpr int N = 768;
+    constexpr int K = 4096;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
diff --git a/test/grouped_gemm/test_grouped_gemm_util.hpp b/test/grouped_gemm/test_grouped_gemm_util.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..b61118b5120e577defe9453c741875cafe33d168
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_util.hpp
@@ -0,0 +1,249 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <array>
+#include <string>
+#include <sstream>
+#include <tuple>
+#include <vector>
+#include <gtest/gtest.h>
+
+#include "ck/ck.hpp"
+#include "ck/stream_config.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/sequence.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/utility/number.hpp"
+#include "profiler/profile_grouped_gemm_impl.hpp"
+
+namespace ck {
+namespace test {
+
+template <typename Range>
+std::string serialize_range(const Range& range)
+{
+    std::stringstream ss;
+    for(auto& r : range)
+    {
+        ss << r << ", ";
+    }
+    std::string str = ss.str();
+    return std::string(str.begin(), str.end() - 2);
+}
+
+template <typename Tuple>
+class TestGroupedGemm : public testing::TestWithParam<int>
+{
+    protected:
+    using ALayout   = std::tuple_element_t<0, Tuple>;
+    using BLayout   = std::tuple_element_t<1, Tuple>;
+    using ELayout   = std::tuple_element_t<2, Tuple>;
+    using ADataType = std::tuple_element_t<3, Tuple>;
+    using BDataType = std::tuple_element_t<4, Tuple>;
+    using EDataType = std::tuple_element_t<5, Tuple>;
+
+    public:
+    static constexpr bool verify_     = true;
+    static constexpr int init_method_ = 1; // decimal value initialization
+    static constexpr bool log_        = false;
+    static constexpr bool bench_      = false; // measure kernel performance
+
+    void SetUp() override {}
+
+    void Run(const std::vector<int>& Ms,
+             const std::vector<int>& Ns,
+             const std::vector<int>& Ks,
+             const std::vector<int>& StrideAs,
+             const std::vector<int>& StrideBs,
+             const std::vector<int>& StrideCs,
+             int kbatch = 1)
+    {
+        bool pass = ck::profiler::profile_grouped_gemm_impl<ADataType,
+                                                            BDataType,
+                                                            EDataType,
+                                                            float,
+                                                            ALayout,
+                                                            BLayout,
+                                                            ELayout>(
+            verify_, init_method_, log_, bench_, Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch);
+        EXPECT_TRUE(pass);
+    }
+};
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          tensor_operation::device::GemmSpecialization GemmSpec,
+          ck::index_t KPerBlock,
+          ck::index_t K1,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock>
+struct DeviceGroupedGemmSplitkInstanceWrapper
+{
+    using F16         = half_t;
+    using F32         = float;
+    using Row         = ck::tensor_layout::gemm::RowMajor;
+    using Col         = ck::tensor_layout::gemm::ColumnMajor;
+    using PassThrough = tensor_operation::element_wise::PassThrough;
+
+    using EmptyTuple = ck::Tuple<>;
+
+    template <ck::index_t... Is>
+    using S = ck::Sequence<Is...>;
+
+    template <ck::index_t N>
+    using I = ck::Number<N>;
+
+    using ABlockTransferThreadClusterArrageOrder =
+        std::conditional_t<std::is_same_v<ALayout, Row>, S<0, 2, 1, 3>, S<0, 1, 3, 2>>;
+    using ABlockTransferSrcAccessOrder =
+        std::conditional_t<std::is_same_v<ALayout, Row>, S<0, 2, 1, 3>, S<0, 1, 3, 2>>;
+    using ABlockTransferSrcVectorDim = std::conditional_t<std::is_same_v<ALayout, Row>, I<3>, I<2>>;
+    using ABlockTransferDstScalarPerVector_K1 =
+        std::conditional_t<std::is_same_v<ALayout, Row>, I<8>, I<2>>;
+    using ABlockLdsAddExtraM = std::conditional_t<std::is_same_v<ALayout, Row>, I<1>, I<0>>;
+
+    using BBlockTransferThreadClusterArrageOrder =
+        std::conditional_t<std::is_same_v<BLayout, Row>, S<0, 1, 3, 2>, S<0, 2, 1, 3>>;
+    using BBlockTransferSrcAccessOrder =
+        std::conditional_t<std::is_same_v<BLayout, Row>, S<0, 1, 3, 2>, S<0, 2, 1, 3>>;
+    using BBlockTransferSrcVectorDim = std::conditional_t<std::is_same_v<BLayout, Row>, I<2>, I<3>>;
+    using BBlockTransferDstScalarPerVector_K1 =
+        std::conditional_t<std::is_same_v<ALayout, Row>, I<2>, I<8>>;
+    using BBlockLdsAddExtraM = std::conditional_t<std::is_same_v<ALayout, Row>, I<0>, I<1>>;
+
+    using DeviceGroupedGemmSplitKInstance =
+        tensor_operation::device::DeviceGroupedGemmXdlSplitKCShuffle<
+            ALayout,
+            BLayout,
+            EmptyTuple,
+            ELayout,
+            F16,
+            F16,
+            F32,
+            F16,
+            EmptyTuple,
+            F16,
+            PassThrough,
+            PassThrough,
+            PassThrough,
+            GemmSpec,
+            1,
+            128,
+            128,
+            128,
+            KPerBlock,
+            K1,
+            K1,
+            32,
+            32,
+            4,
+            2,
+            S<1, 4, 32, 1>,
+            ABlockTransferThreadClusterArrageOrder,
+            ABlockTransferSrcAccessOrder,
+            ABlockTransferSrcVectorDim::value,
+            ABlockTransferSrcScalarPerVector,
+            ABlockTransferDstScalarPerVector_K1::value,
+            ABlockLdsAddExtraM::value,
+            S<1, 4, 32, 1>,
+            BBlockTransferThreadClusterArrageOrder,
+            BBlockTransferSrcAccessOrder,
+            BBlockTransferSrcVectorDim::value,
+            BBlockTransferSrcScalarPerVector,
+            BBlockTransferDstScalarPerVector_K1::value,
+            BBlockLdsAddExtraM::value,
+            1,
+            1,
+            S<1, 16, 1, 8>,
+            CDEBlockTransferScalarPerVector_NPerBlock>;
+
+    bool IsSupported(const std::vector<int>& Ms,
+                     const std::vector<int>& Ns,
+                     const std::vector<int>& Ks,
+                     const std::vector<int>& StrideAs,
+                     const std::vector<int>& StrideBs,
+                     const std::vector<int>& StrideCs,
+                     int kbatch = 1) const
+    {
+        std::size_t n_groups = Ms.size();
+        EXPECT_TRUE(Ns.size() == n_groups && Ks.size() == n_groups && StrideAs.size() == n_groups &&
+                    StrideBs.size() == n_groups && StrideCs.size() == n_groups)
+            << "The number of groups is not consistent!";
+
+        std::vector<tensor_operation::device::GemmDesc> gemm_descs;
+
+        for(std::size_t i = 0; i < n_groups; ++i)
+        {
+            gemm_descs.push_back(tensor_operation::device::GemmDesc{
+                Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
+        }
+
+        std::vector<const void*> p_As(n_groups, nullptr);
+        std::vector<const void*> p_Bs(n_groups, nullptr);
+        std::vector<void*> p_Cs(n_groups, nullptr);
+        auto p_Ds = std::vector<std::array<const void*, 0>>{};
+
+        auto ggemm_instance = DeviceGroupedGemmSplitKInstance{};
+        auto argument       = ggemm_instance.MakeArgument(
+            p_As, p_Bs, p_Ds, p_Cs, gemm_descs, PassThrough{}, PassThrough{}, PassThrough{});
+        if(kbatch > 1)
+        {
+            ggemm_instance.SetKBatchSize(argument, kbatch);
+        }
+
+        return ggemm_instance.IsSupportedArgument(argument);
+    }
+
+    float Run(const std::vector<int>& Ms,
+              const std::vector<int>& Ns,
+              const std::vector<int>& Ks,
+              const std::vector<int>& StrideAs,
+              const std::vector<int>& StrideBs,
+              const std::vector<int>& StrideCs,
+              int kbatch = 1) const
+    {
+        std::size_t n_groups = Ms.size();
+        EXPECT_TRUE(Ns.size() == n_groups && Ks.size() == n_groups && StrideAs.size() == n_groups &&
+                    StrideBs.size() == n_groups && StrideCs.size() == n_groups)
+            << "The number of groups is not consistent!";
+
+        std::vector<tensor_operation::device::GemmDesc> gemm_descs;
+
+        for(std::size_t i = 0; i < n_groups; ++i)
+        {
+            gemm_descs.push_back(tensor_operation::device::GemmDesc{
+                Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
+        }
+
+        std::vector<const void*> p_As(n_groups, nullptr);
+        std::vector<const void*> p_Bs(n_groups, nullptr);
+        std::vector<void*> p_Cs(n_groups, nullptr);
+        auto p_Ds = std::vector<std::array<const void*, 0>>{};
+
+        auto ggemm_instance = DeviceGroupedGemmSplitKInstance{};
+        auto argument       = ggemm_instance.MakeArgument(
+            p_As, p_Bs, p_Ds, p_Cs, gemm_descs, PassThrough{}, PassThrough{}, PassThrough{});
+        if(kbatch > 1)
+        {
+            ggemm_instance.SetKBatchSize(argument, kbatch);
+        }
+
+        EXPECT_TRUE(ggemm_instance.IsSupportedArgument(argument));
+        auto invoker = ggemm_instance.MakeInvoker();
+        DeviceMem gemm_desc_workspace(ggemm_instance.GetWorkSpaceSize(&argument));
+        ggemm_instance.SetWorkSpacePointer(&argument, gemm_desc_workspace.GetDeviceBuffer());
+        return invoker.Run(argument, StreamConfig{nullptr, false});
+    }
+};
+
+} // namespace test
+} // namespace ck
diff --git a/test/magic_number_division/magic_number_division.cpp b/test/magic_number_division/magic_number_division.cpp
index 680fddf1933611dc3088cb673ece0dba9e1f3911..253f21e91f80c96c081108b47b045c3902b36926 100644
--- a/test/magic_number_division/magic_number_division.cpp
+++ b/test/magic_number_division/magic_number_division.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/test/normalization/CMakeLists.txt b/test/normalization/CMakeLists.txt
index 456423f25d20c04f21a6c3dab85b49afdac077dd..a5d7fb2982c9dac7cc2da58a8f9f6be413cfbd18 100644
--- a/test/normalization/CMakeLists.txt
+++ b/test/normalization/CMakeLists.txt
@@ -1,17 +1,16 @@
-add_custom_target(test_layernorm)
+add_custom_target(test_normalization)
 
 add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
 add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
 add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
-add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp) 
-
+add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
 
 target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance)
 target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance)
 target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
 target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
 
-add_dependencies(test_layernorm test_layernorm2d_fp32)
-add_dependencies(test_layernorm test_layernorm2d_fp16)
-add_dependencies(test_layernorm test_groupnorm_fp16)
-add_dependencies(test_layernorm test_groupnorm_fp32)
+add_dependencies(test_normalization test_layernorm2d_fp32)
+add_dependencies(test_normalization test_layernorm2d_fp16)
+add_dependencies(test_normalization test_groupnorm_fp16)
+add_dependencies(test_normalization test_groupnorm_fp32)
diff --git a/test/normalization/test_groupnorm_fp16.cpp b/test/normalization/test_groupnorm_fp16.cpp
index 636e522dce3742d801eb256d35e43d49832d4878..325ea75fe5668897a5b489216db042180b14a8ef 100644
--- a/test/normalization/test_groupnorm_fp16.cpp
+++ b/test/normalization/test_groupnorm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_groupnorm_impl.hpp"
@@ -12,11 +12,11 @@ template <typename Tuple>
 class TestGroupnorm : public ::testing::Test
 {
     protected:
-    using XDataType     = std::tuple_element_t<0, Tuple>;
-    using GammaDataType = std::tuple_element_t<1, Tuple>;
-    using BetaDataType  = std::tuple_element_t<2, Tuple>;
-    using AccDataType   = std::tuple_element_t<3, Tuple>;
-    using YDataType     = std::tuple_element_t<4, Tuple>;
+    using XDataType       = std::tuple_element_t<0, Tuple>;
+    using GammaDataType   = std::tuple_element_t<1, Tuple>;
+    using BetaDataType    = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType = std::tuple_element_t<3, Tuple>;
+    using YDataType       = std::tuple_element_t<4, Tuple>;
 
     void Run()
     {
@@ -36,7 +36,7 @@ class TestGroupnorm : public ::testing::Test
                 ck::profiler::profile_groupnorm_impl<XDataType,
                                                      GammaDataType,
                                                      BetaDataType,
-                                                     AccDataType,
+                                                     ComputeDataType,
                                                      YDataType>(true, 2, false, false, length);
             EXPECT_TRUE(success);
         }
@@ -44,7 +44,7 @@ class TestGroupnorm : public ::testing::Test
 };
 
 using KernelTypes = ::testing::Types<
-    // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
+    // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
     std::tuple<F16, F16, F16, F32, F16>>;
 
 TYPED_TEST_SUITE(TestGroupnorm, KernelTypes);
diff --git a/test/normalization/test_groupnorm_fp32.cpp b/test/normalization/test_groupnorm_fp32.cpp
index ef492664bff55fe9a1566ef9736e89dbd224f259..ec88442fc02623767f299de36f8b6c5810569cc6 100644
--- a/test/normalization/test_groupnorm_fp32.cpp
+++ b/test/normalization/test_groupnorm_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_groupnorm_impl.hpp"
@@ -12,11 +12,11 @@ template <typename Tuple>
 class TestGroupnorm : public ::testing::Test
 {
     protected:
-    using XDataType     = std::tuple_element_t<0, Tuple>;
-    using GammaDataType = std::tuple_element_t<1, Tuple>;
-    using BetaDataType  = std::tuple_element_t<2, Tuple>;
-    using AccDataType   = std::tuple_element_t<3, Tuple>;
-    using YDataType     = std::tuple_element_t<4, Tuple>;
+    using XDataType       = std::tuple_element_t<0, Tuple>;
+    using GammaDataType   = std::tuple_element_t<1, Tuple>;
+    using BetaDataType    = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType = std::tuple_element_t<3, Tuple>;
+    using YDataType       = std::tuple_element_t<4, Tuple>;
 
     void Run()
     {
@@ -34,7 +34,7 @@ class TestGroupnorm : public ::testing::Test
                 ck::profiler::profile_groupnorm_impl<XDataType,
                                                      GammaDataType,
                                                      BetaDataType,
-                                                     AccDataType,
+                                                     ComputeDataType,
                                                      YDataType>(true, 2, false, false, length);
             EXPECT_TRUE(success);
         }
@@ -42,7 +42,7 @@ class TestGroupnorm : public ::testing::Test
 };
 
 using KernelTypes = ::testing::Types<
-    // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
+    // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
     std::tuple<F32, F32, F32, F32, F32>>;
 
 TYPED_TEST_SUITE(TestGroupnorm, KernelTypes);
diff --git a/test/normalization/test_layernorm2d_fp16.cpp b/test/normalization/test_layernorm2d_fp16.cpp
index eeb8ec150ac63e12c33446f5a5b4f03d2afc0d2f..2222740fcceaa4b9dfe22a299dc0cd3e054e5d35 100644
--- a/test/normalization/test_layernorm2d_fp16.cpp
+++ b/test/normalization/test_layernorm2d_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_layernorm_impl.hpp"
@@ -12,11 +12,11 @@ template <typename Tuple>
 class TestLayernorm2d : public ::testing::Test
 {
     protected:
-    using XDataType     = std::tuple_element_t<0, Tuple>;
-    using GammaDataType = std::tuple_element_t<1, Tuple>;
-    using BetaDataType  = std::tuple_element_t<2, Tuple>;
-    using AccDataType   = std::tuple_element_t<3, Tuple>;
-    using YDataType     = std::tuple_element_t<4, Tuple>;
+    using XDataType       = std::tuple_element_t<0, Tuple>;
+    using GammaDataType   = std::tuple_element_t<1, Tuple>;
+    using BetaDataType    = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType = std::tuple_element_t<3, Tuple>;
+    using YDataType       = std::tuple_element_t<4, Tuple>;
 
     void Run()
     {
@@ -29,7 +29,7 @@ class TestLayernorm2d : public ::testing::Test
             bool success = ck::profiler::profile_layernorm_impl<XDataType,
                                                                 GammaDataType,
                                                                 BetaDataType,
-                                                                AccDataType,
+                                                                ComputeDataType,
                                                                 YDataType,
                                                                 2>(true, 2, false, false, length);
             EXPECT_TRUE(success);
@@ -38,7 +38,7 @@ class TestLayernorm2d : public ::testing::Test
 };
 
 using KernelTypes = ::testing::Types<
-    // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
+    // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
     std::tuple<F16, F16, F16, F32, F16>>;
 
 TYPED_TEST_SUITE(TestLayernorm2d, KernelTypes);
diff --git a/test/normalization/test_layernorm2d_fp32.cpp b/test/normalization/test_layernorm2d_fp32.cpp
index f555b42592a04de8c883db0ab6609e251877e066..30fbe06c60d8ebfcbdd87b3e27c11c601d81635b 100644
--- a/test/normalization/test_layernorm2d_fp32.cpp
+++ b/test/normalization/test_layernorm2d_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_layernorm_impl.hpp"
@@ -12,11 +12,11 @@ template <typename Tuple>
 class TestLayernorm2d : public ::testing::Test
 {
     protected:
-    using XDataType     = std::tuple_element_t<0, Tuple>;
-    using GammaDataType = std::tuple_element_t<1, Tuple>;
-    using BetaDataType  = std::tuple_element_t<2, Tuple>;
-    using AccDataType   = std::tuple_element_t<3, Tuple>;
-    using YDataType     = std::tuple_element_t<4, Tuple>;
+    using XDataType       = std::tuple_element_t<0, Tuple>;
+    using GammaDataType   = std::tuple_element_t<1, Tuple>;
+    using BetaDataType    = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType = std::tuple_element_t<3, Tuple>;
+    using YDataType       = std::tuple_element_t<4, Tuple>;
 
     void Run()
     {
@@ -29,7 +29,7 @@ class TestLayernorm2d : public ::testing::Test
             bool success = ck::profiler::profile_layernorm_impl<XDataType,
                                                                 GammaDataType,
                                                                 BetaDataType,
-                                                                AccDataType,
+                                                                ComputeDataType,
                                                                 YDataType,
                                                                 2>(true, 2, false, false, length);
             EXPECT_TRUE(success);
@@ -38,7 +38,7 @@ class TestLayernorm2d : public ::testing::Test
 };
 
 using KernelTypes = ::testing::Types<
-    // XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
+    // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
     std::tuple<F32, F32, F32, F32, F32>>;
 
 TYPED_TEST_SUITE(TestLayernorm2d, KernelTypes);
diff --git a/test/pool_fwd/CMakeLists.txt b/test/pool_fwd/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6f59b95f6fce8679a022f2bb57de80983cf261a5
--- /dev/null
+++ b/test/pool_fwd/CMakeLists.txt
@@ -0,0 +1,16 @@
+add_custom_target(test_pool_fwd)
+
+add_gtest_executable(test_avg_pool2d_fwd test_avg_pool2d_fwd.cpp)
+add_gtest_executable(test_avg_pool3d_fwd test_avg_pool3d_fwd.cpp)
+add_gtest_executable(test_max_pool2d_fwd test_max_pool2d_fwd.cpp)
+add_gtest_executable(test_max_pool3d_fwd test_max_pool3d_fwd.cpp)
+
+target_link_libraries(test_avg_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
+target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
+target_link_libraries(test_max_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
+target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
+
+add_dependencies(test_pool_fwd test_avg_pool2d_fwd)
+add_dependencies(test_pool_fwd test_avg_pool3d_fwd)
+add_dependencies(test_pool_fwd test_max_pool2d_fwd)
+add_dependencies(test_pool_fwd test_max_pool3d_fwd)
diff --git a/test/pool_fwd/test_avg_pool2d_fwd.cpp b/test/pool_fwd/test_avg_pool2d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..72749fd6e1563f9b432a61871982bc7f6a2e9661
--- /dev/null
+++ b/test/pool_fwd/test_avg_pool2d_fwd.cpp
@@ -0,0 +1,56 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool2d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestAvgPool2dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            bool success =
+                ck::profiler::profile_pool2d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::AVG,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestAvgPool2dFwd, KernelTypes);
+TYPED_TEST(TestAvgPool2dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_avg_pool3d_fwd.cpp b/test/pool_fwd/test_avg_pool3d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..00cc3740fa5cc775d43bf60300b693789c7e2d4d
--- /dev/null
+++ b/test/pool_fwd/test_avg_pool3d_fwd.cpp
@@ -0,0 +1,56 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool3d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestAvgPool3dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            bool success =
+                ck::profiler::profile_pool3d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::AVG,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestAvgPool3dFwd, KernelTypes);
+TYPED_TEST(TestAvgPool3dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_max_pool2d_fwd.cpp b/test/pool_fwd/test_max_pool2d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1cf1314f4682acb528a30013011b1cdb21ea965f
--- /dev/null
+++ b/test/pool_fwd/test_max_pool2d_fwd.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool2d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestMaxPool2dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            // max pool
+            bool success =
+                ck::profiler::profile_pool2d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::MAX,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+
+            // max pool + index
+            success = ck::profiler::profile_pool2d_fwd_impl<InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ck::ReduceTensorOp::MAX,
+                                                            false,
+                                                            true>(true,
+                                                                  2,
+                                                                  false,
+                                                                  false,
+                                                                  param.length_,
+                                                                  param.window_spatial_lengths_,
+                                                                  param.window_strides_,
+                                                                  param.input_left_pads_,
+                                                                  param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestMaxPool2dFwd, KernelTypes);
+TYPED_TEST(TestMaxPool2dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_max_pool3d_fwd.cpp b/test/pool_fwd/test_max_pool3d_fwd.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..0b0de4d90ec57aa57db1d8a173d419f879a1ea87
--- /dev/null
+++ b/test/pool_fwd/test_max_pool3d_fwd.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool3d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestMaxPool3dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            // max pool
+            bool success =
+                ck::profiler::profile_pool3d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::MAX,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+
+            // max pool + index
+            success = ck::profiler::profile_pool3d_fwd_impl<InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ck::ReduceTensorOp::MAX,
+                                                            false,
+                                                            true>(true,
+                                                                  2,
+                                                                  false,
+                                                                  false,
+                                                                  param.length_,
+                                                                  param.window_spatial_lengths_,
+                                                                  param.window_strides_,
+                                                                  param.input_left_pads_,
+                                                                  param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestMaxPool3dFwd, KernelTypes);
+TYPED_TEST(TestMaxPool3dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_pool_fwd_common.hpp b/test/pool_fwd/test_pool_fwd_common.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..f018635170d9b3eb2235563fd04faefec38fea11
--- /dev/null
+++ b/test/pool_fwd/test_pool_fwd_common.hpp
@@ -0,0 +1,31 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "ck/ck.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+using I32 = int32_t;
+using ck::index_t;
+
+struct PoolingParam
+{
+    PoolingParam(const std::vector<index_t>& length,
+                 const std::vector<index_t>& window_spatial_lengths,
+                 const std::vector<index_t>& window_strides,
+                 const std::vector<index_t>& input_left_pads,
+                 const std::vector<index_t>& input_right_pads)
+        : length_(length),
+          window_spatial_lengths_(window_spatial_lengths),
+          window_strides_(window_strides),
+          input_left_pads_(input_left_pads),
+          input_right_pads_(input_right_pads)
+    {
+    }
+    std::vector<index_t> length_;
+    std::vector<index_t> window_spatial_lengths_;
+    std::vector<index_t> window_strides_;
+    std::vector<index_t> input_left_pads_;
+    std::vector<index_t> input_right_pads_;
+};
diff --git a/test/reduce/reduce_no_index.cpp b/test/reduce/reduce_no_index.cpp
index 3f4d0676b4da84751dd74cd4e76266059c056d67..1ab452442d0b219820e208420a4c4c04a019e9a1 100644
--- a/test/reduce/reduce_no_index.cpp
+++ b/test/reduce/reduce_no_index.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <getopt.h>
 
diff --git a/test/reduce/reduce_with_index.cpp b/test/reduce/reduce_with_index.cpp
index c616a68e741e14119d70e0d1d4ba508e371dea78..0301669c5a7578b7c6a0c6ba160685b13295b492 100644
--- a/test/reduce/reduce_with_index.cpp
+++ b/test/reduce/reduce_with_index.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <getopt.h>
 
diff --git a/test/reference_conv_fwd/reference_conv_fwd.cpp b/test/reference_conv_fwd/reference_conv_fwd.cpp
index 1f9ba0064cb3b22c394c75d5119f71a188d11a5a..b3328e4b365268af641a6ecb1a65ded872f94ddd 100644
--- a/test/reference_conv_fwd/reference_conv_fwd.cpp
+++ b/test/reference_conv_fwd/reference_conv_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cmath>
 #include <cstdlib>
diff --git a/test/softmax/test_softmax_interface.cpp b/test/softmax/test_softmax_interface.cpp
index 8cac0ba0f52b035dae2fb52f0b0346d3267e5b7b..25f666f0ea99eb12e0fa248fdd573aebed0de7a6 100644
--- a/test/softmax/test_softmax_interface.cpp
+++ b/test/softmax/test_softmax_interface.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <stdexcept>
 #include <vector>
diff --git a/test/softmax/test_softmax_rank3.cpp b/test/softmax/test_softmax_rank3.cpp
index 5691ee3f6cd5be1bd8ff1075cb57fcfd86aa8322..43ae11bf1f243e930517e618b167af30e1599e0d 100644
--- a/test/softmax/test_softmax_rank3.cpp
+++ b/test/softmax/test_softmax_rank3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <stdexcept>
@@ -13,7 +13,6 @@ using I = ck::Number<N>;
 
 using F16 = ck::half_t;
 using F32 = float;
-using I8  = int8_t;
 
 template <typename Tuple>
 class TestSoftmax : public ck::TestSoftmax<Tuple>
@@ -24,8 +23,7 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
     std::tuple<       F16,         F32,         F16,    I<3>>,
-    std::tuple<       F32,         F32,         F32,    I<3>>,
-    std::tuple<        I8,         F32,          I8,    I<3>>
+    std::tuple<       F32,         F32,         F32,    I<3>>
     >;
 // clang-format on
 
diff --git a/test/softmax/test_softmax_rank4.cpp b/test/softmax/test_softmax_rank4.cpp
index f0b22df25ebdfaaecbf4548ec2b44fdcd477aae4..5cf96bbaa8553ab404693f555a776d22fa37f9e4 100644
--- a/test/softmax/test_softmax_rank4.cpp
+++ b/test/softmax/test_softmax_rank4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <stdexcept>
@@ -13,7 +13,6 @@ using I = ck::Number<N>;
 
 using F16 = ck::half_t;
 using F32 = float;
-using I8  = int8_t;
 
 template <typename Tuple>
 class TestSoftmax : public ck::TestSoftmax<Tuple>
@@ -24,8 +23,7 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
     std::tuple<       F16,         F32,         F16,    I<4>>,
-    std::tuple<       F32,         F32,         F32,    I<4>>,
-    std::tuple<        I8,         F32,          I8,    I<4>>
+    std::tuple<       F32,         F32,         F32,    I<4>>
     >;
 // clang-format on
 
diff --git a/test/softmax/test_softmax_util.hpp b/test/softmax/test_softmax_util.hpp
index 40b300cf9927de3556c586805923de9ab97db466..1409af8453b3a9d722fd5074d311f7066f295c65 100644
--- a/test/softmax/test_softmax_util.hpp
+++ b/test/softmax/test_softmax_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -61,8 +61,92 @@ class TestSoftmax : public ::testing::Test
         int init_method = 1; // integer value initialization
         bool log        = false;
         std::vector<ck::index_t> strides; // intenionally empty, to get packed layout.
-        bool pass = ck::profiler::profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank>(
-            verify_, init_method, log, bench_, in_length, strides, reduce_dims, alpha, beta);
+        bool pass = false;
+
+        if constexpr(Rank == 3)
+        {
+            if(reduce_dims.size() == 1)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 1>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 2)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 2>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 3)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 3>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+        }
+        else if constexpr(Rank == 4)
+        {
+            if(reduce_dims.size() == 1)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 1>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 2)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 2>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 3)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 3>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 4)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 4>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+        };
+
         EXPECT_TRUE(pass);
     }
 
diff --git a/test/space_filling_curve/space_filling_curve.cpp b/test/space_filling_curve/space_filling_curve.cpp
index c7f6759e819512f55c17463ffdb503e55bb8c59e..a192ecb28f1b6c67fa7ec27790349e041c6718fb 100644
--- a/test/space_filling_curve/space_filling_curve.cpp
+++ b/test/space_filling_curve/space_filling_curve.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 #include <iostream>
diff --git a/test/wmma_op/wmma_op.cpp b/test/wmma_op/wmma_op.cpp
index 761c15f1dd8a9d22de14a2998d73a3877b1f3003..47d8c7ed6f35236192cc9c4cfacde2e754560241 100644
--- a/test/wmma_op/wmma_op.cpp
+++ b/test/wmma_op/wmma_op.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/wmma_op/wmma_op_util.hpp b/test/wmma_op/wmma_op_util.hpp
index ef3f831abdea2252f19e1097d2c6a7252bc19b19..49782bce6e21d4a250b0d07526abf4e39ac57b9d 100644
--- a/test/wmma_op/wmma_op_util.hpp
+++ b/test/wmma_op/wmma_op_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -97,6 +97,7 @@ builtin_wmma_naive_selector<int4x16_t,
 template <typename src_t, typename dst_t, typename acc_t, index_t acc_num>
 __global__ void matmul(const src_t* a, const src_t* b, dst_t* c)
 {
+    __shared__ src_t p_shared[16 * 16 * 2];
     const int lIdx = threadIdx.x;
     // a and b fragments are stored in 8 VGPRs each, in packed format, so 16 elements each for a and
     // b a_frag will store one column of the 16x16 matrix tile b_frag will store one row of the
@@ -104,6 +105,9 @@ __global__ void matmul(const src_t* a, const src_t* b, dst_t* c)
     using src_vec  = typename vector_type<src_t, 16>::type;
     src_vec a_frag = {};
     src_vec b_frag = {};
+
+    src_vec a_temp = {};
+    src_vec b_temp = {};
     // initialize c fragment to 0
     using acc_vec = StaticBufferTupleOfVector<AddressSpaceEnum::Vgpr, acc_t, 1, acc_num, true>;
     acc_vec c_thread_buf_;
@@ -111,21 +115,57 @@ __global__ void matmul(const src_t* a, const src_t* b, dst_t* c)
     // lane is (0-31) mod 16 instead of 0-31 due to matrix replication in gfx11
     // see https://atlvsp3.amd.com/sp3_gfx11_5_instructions.pdf page 482
     // TODO: remove this dependency in gfx12 https://ontrack-internal.amd.com/browse/DEGFXSP3-101
-    const int lane = lIdx % 16;
+    const int lane    = lIdx % 16;
+    const int lane_lo = lIdx / 2;
+    const int lane_hi = lIdx % 2;
+    for(int ele = 0; ele < 8; ++ele)
+    {
+        a_temp[ele] = a[8 * lane_hi + 16 * lane_lo + ele];
+    }
+
+    for(int ele = 0; ele < 8; ++ele)
+    {
+        b_temp[ele] = b[8 * lane_hi + 16 * lane_lo + ele];
+    }
+
+    __syncthreads();
+
+    for(int ele = 0; ele < 8; ++ele)
+    {
+        p_shared[8 * 16 * lane_hi + 8 * lane_lo + ele] = a_temp[ele];
+    }
+
+    for(int ele = 0; ele < 8; ++ele)
+    {
+        p_shared[8 * 16 * lane_hi + 8 * lane_lo + ele + 16 * 16] = b_temp[ele];
+    }
+
+    asm volatile("\
+    s_waitcnt lgkmcnt(0) \n \
+    s_barrier \
+    " ::);
 
     for(int ele = 0; ele < 16; ++ele)
     {
-        b_frag[ele] = b[16 * lane + ele];
+        b_frag[ele] = p_shared[(ele / 8) * 16 * 8 + 8 * lane + ele % 8 + 16 * 16];
     }
     // follow origin design
     for(int ele = 0; ele < 16; ++ele)
     {
-        a_frag[ele] = a[16 * lane + ele];
+        a_frag[ele] = p_shared[(ele / 8) * 16 * 8 + 8 * lane + ele % 8];
     }
 
+    asm volatile("\
+    s_waitcnt lgkmcnt(0) \n \
+    s_barrier \
+    " ::);
+
     // sync threads, similar to mma_sync
-    __syncthreads();
+    // __syncthreads();
     builtin_wmma_naive_selector<src_vec, acc_vec>(a_frag, b_frag, c_thread_buf_);
+    // since only fp16_fp32 asm wmma implemented for experiment purpose, restrict test case to fp16
+    // when enable this ck::amd_assembly_wmma_f32_16x16x16_f16_w32(a_frag, b_frag,
+    // c_thread_buf_.GetVectorTypeReference(Number<0>{}).template AsType<float8_t>()(Number<0>{}));
     __syncthreads();
     // wait for results, similar to mma_sync
     static_for<0, 8, 1>{}([&](auto ele) {